Grand Challenges For The Next 20 Years

terrapyn writes "Infoworld is reporting: 'A group of British computer scientists have proposed a number of grand challenges for IT that they hope will drive forward research, similar to the way the human genome project drove life sciences research through the 1990s.' Did they get it right? What are some other worthy computing challenges?"

Comments

Just ONE request...

[Thunderstruck]

A battery, a really good battery. Something that'll make my laptop last as long as my Palm. Or maybe power a light-saber... But really all we need for our dreams to come true is a good battery.

Re:Just ONE request...

[chris09876]

True.. that's one thing they left off their list. Battery life hasn't increased at the rate as I'd like it to. Id would be a beautiful thing if I only had to charge my PDA once/month, or my laptop could go a week without charging

Re:Just ONE request...

[Anonymous Coward]

This is not a request for IT. INFORMATION TECHNOLOGY.

It is a request for ET Engineering Technology. .segmond

Re:Just ONE request...

[iMaple]

My palm lasts for ever and so does my laptop though the laptop is more fun, but neither requires batteries, the mail order did not have a battery operated auto model ... ooops you meant computers

Re:Just ONE request...

[plover]

Fuel cells.

They already have prototypes small enough to power a cell phone, and they're approaching the marketplace. Cost is unknown, but you can expect them to be expensive at first. And if they take platinum as a catalyst, costs will of course stay high.

It will remain to be seen if people will accept carrying volatile fluids around with them, but I'm betting they'll come out with a "clean change" cartridge system that people will like. Just think: no recharging time. A small reservoir will probably allow for a hot-swap of the cartridge as well, meaning not even any down-time.

Next problem?

Re:Just ONE request...

[mboverload]

Mobile power (batteries) is the only thing restricting us from having amazing portable machines. Lion tech is getting old and unable to power our society. With processor speeds reaching 4ghz soon, the battery "industry" is lagging way behind. Hell, we had like 386's when Lithium ion came out.

Re:Just ONE request...

[danheskett]

What we really need someone to do is design a laptop from the ground-up for maximum battery life. And I don't just mean the processor.

Look at every function, every component, and remove/reduce everything unnecessary. Combine anything possible. Strip it down to nothing. Give me an old school screen. And I mean that. I don't even care if its an old gas plasma or monochrome display.

I am talking about a laptop that will run with average use for 2 weeks between charges. I don't even care if it's a refined 486/25. Whatever it's speed/capabilities, I'll find an OS to run with it. If that means a console only Linux distro, I'm fine with that.

The fact that you can't get a laptop that can truly run for more than 8 hrs off of battery without insane power saving options is nuts.

Give me a 1 lb 1/2 inch think laptop with a low-power "486 level" processor, minimal graphics card, 1 gb flash card instead of a harddisk, optical disk and wireless network adapter. I don't need no stinking parallel port, no freaking COM ports, S-Video out, no ability to display two video displays at once, no freaking docking ports, maybe a USB port if its not too much trouble, no firewire, no infrared, no onboard ethernet, no onboard modem and definately no line input/microphone jacks. I don't need to no high-powered speakers, a head-phone jack and 16-bit 2-channel stereo sound will do just fine. A 10" screen will do fine, an optional mechanical backlight switch will do fine. Color is nice if you can have it, otherwise, give me a 256-shades of gray and 800x600 resolution.

If someone could persuade a hardware manufacturing plant to make a non-name version of this laptop I am sure it would sell. With the best battery you can buy and this unit you could probabl sell them retail for $399 and make a decent profit. If you can get a CPU down to a low-enough voltage and wattage - and it doesn't have to be an x86 processor mind you - I could see a life of 24 hrs continuous being plausable. Whatever you can get that's low wattage (4 Watts? What's reasonable? My P4 takes an insane amount.. what, 90 Watts all by itself, without anything else in the box? pfft).

Enough of this diatribe. This should be a no brainer. You could sell millions of these units easily. Put together a nice tightly integrated suite of tools - simple e-mail, simple web-browser, simple office suite, etc and you'll be making millions.

Here's an IT challange...

[bulliver]

Make Windows secure.

Nothing new here

[MAdMaxOr]

I didn't see anything that hasn't already been proposed many times before. Also, the article was short, and the descriptions were very general and boring.

**yawn**

Re:Nothing new here

[SupremeTaco]

Yes, the descriptions are vague, and I think necessarily so. It's a challenge to possibly develop new technologies that will do these things, or perhaps make them obselete or un-needed. Also, sometimes the end result is boring, but the technology needed to get there is pretty exciting. A lot of people are bored now when you talk about putting a satellite in orbit, or exploring the bottom of the ocean, but when you start to break down the technology that it takes to make it there, you kinda go "WOW!"

Setting the goal is the easy part. Making it happen should be fun.

Who knows what will happen

[chris09876]

They're setting these as goals for the next 15 years... but who really knows what's going to happen 15 years from now? If Moore's law holds (and we have no reason to think it won't), we'll have almost 2^10 times the computing power we do today. That's a huge number!! Setting these goals is a nice idea..., but who knows what the world has in store.

Re:Who knows what will happen

[Alrua]

Moore's Law doesn't apply to clock speed, but to the number of transistors in microchips. The number of transistors continues to rise exponentially like Moore predicted...

Re:Who knows what will happen

[rbarreira]

No surprise considering that applications are getting heavier and heavier... Most programmers no longer care about optimizing their code, as they used to (and had to) some years ago.

Re:Who knows what will happen

[psetzer]

Processing power doesn't drive innovation as some claim a lack of it drives efficiency. Even if, in fifteen years, we've got computers with a thousand times the circuitry, programs will run just as fast as they do today, and what we use it for will generally be the same. What innovations do occur seem to be folded back into existing technology, making it better. This looks at innovations that do more than just make a better search engine, and change how we look at the computer.

Ubiquitous computing is an interesting completely transformational view of how we look at computers. Back in the early 1900s, people bought accessories for their electric motors. They were simply too expensive to put a seperate one in each appliance, so if you wanted a vaccuum cleaner, you'd buy the attachment that turned your motor into a vacuum cleaner. Now, you buy something with an electric motor, and odds are that they don't even mention one's in there. The difficult part of ubiquitous computing isn't cheap, powerful computers; that's solved. The difficult part is getting everything to work together and handle stuff that doesn't want to work together nicely. What do you do when you are given some request that you don't know how to handle? Do you ignore it, or do you pass it off to someone who knows how to handle it? If it's something that you've never even heard of before, then would you know who to hand it off to? What you need is some sort of protocol that's expandable, universal, and standardized, and a computing framework that's capable of handling it.

No matter what anybody says, XML isn't sufficient. Objects in the framework need to be capable of broadcasting their capabilities, and other objects need to know how to use those capabilities. It would be nice if we even had that in an individual computer right now. If something needs to show a picture, it should be able to find the program that does that without the user needing to tell it that. Figure that out, and the world will beat a path to your door.

DATA DATA DATA

[Ars-Fartsica]

We are being buried in data and are just beginning to adapt the crudest methods for organizing it and mining it. If in 20 years we have not solved the problem of dealing with giant piles of data, then IT will become a cost instead of a benefit.

Re:DATA DATA DATA

[nine-times]

I'll agree with this one. I look at my company's servers, and it seems like we just keep having to add more hard drives. Some of it's because people are disorganized, but sometimes people are disorganized because of the massive amounts of data that they're dealing with.

I have users with multi-GB mailboxes that can't quite be deleted, but archiving it doesn't really solve the problem either, it just makes it harder for the user to find what he's looking for.

So, it's a basic problem. Every day, we're generating more data. The amount of data (in bytes) is going up every day, as computers are more easily able to deal with higher resolution pictures and movies. But what do we do with all this data? Just keep writing it to tape and storing it in bunkers? After we accrue enough data, what's the point of keeping it?-- you won't be able to find anything anymore.

It's a real problem for me, both as an IT pro and personally. When dealing with so much data, how do you:

1. keep everything you want
2. make it easy to find what you want when you want it
3. make it easy to access what you want when you want it
4. throw away everything you aren't going to want

And how do you do all that with:

1. a solution a non-techie can deal with (grandma needs her data safe, too)
2. security from unauthorized access
3. security from data loss (off-site backups?)
4. an affordable price (both corporate and personal solutions)
5. without spending the amount of time on this that only an obsessive compulsive would consider acceptable

I haven't seen an acceptable solution yet.

Re:DATA DATA DATA

[joshsnow]

I seem to remember, from my university days, being told that meaningful information (as opposed to "data") must be relevant, timely, structured and domain specific.

I agree, we are being buried in data but perhaps that's because the emphasis is on collecting data rather than managing information.

IT will continue to be a benefit so long as we focus on precisely what we're gathering and structuring data for.

Re:DATA DATA DATA

[Linker3000]

I'd actually put it another way - many organisations are only beginning to realise that they need to do something (or can do something) with all the information they are gathering and/or that there is value in what the information can tell them about their customer base (internal or external) and business workings.

To imply that we're only just working out what to do with all our information is not quite right because the principles of Knowledge Management are well established - for example one of the often-quoted books on the subject was published in 1971 by CW Churchman (useful info here). The main problem is getting organisations to 'see the wood for the trees' and to invest some time and funds in analysing the potential in the information they possess - such activities often pay for themselves in a surprisingly short time.

Anyone wanting to read more could do worse than start at brint.com - the web site looks daunting but it's well worth a visit and spending some time there.

Re:DATA DATA DATA

[Cassanova]

My simplistic solution from what I have observed personally, would be two steps:

1. Archive data older than n days to backup/removeable storage, delete backedup data from your main field of view (desktop/folders etc from wherever they came) so you now have a fresh slate to begin with. This ensures less clutter.

2. Destroy backedup data after y days. Yes, quite simply destroy it. Make y sufficiently large that you can always get something back if you really wanted it. Of course y is >> x (significantly larger than). If you want something put in cold storage forever, explicitly move it into a "cold storage archive" in step 1 above. I'm guessing there will be very little stuff deserving this status so storing it will be manageable. This step ensures that unneeded data does not last forever and ever.

A lot of clutter is built up by having the notion that you will "need it *someday*" - thats a fallacy - you mostly end up never touching 80% of your stashed stuff so they can be safely deleted.

Re:DATA DATA DATA

[nine-times]

-- sure you will want to wrap it into some kind of GUI for you grandma ;)

Yeah, but part of my point was, not every grandma has a me to set *anything* up. I don't want to have to build a Unix system and write a custom solution for my grandmother anyway. I am not a one-man full-time tech-support staff for everyone I know. When I talk about a solution, I mean something that comes with the computer or is an easy-to-install add-on that grandma can do herself. I mean something that I can point out to some know-nothing and say "Buy this. It'll take care of your problems."

I suspect that banks will start providing safes for data soon - with some kind access like ssh

For grandma, they'd better have a better interface than CLI SSH. Maybe a program that uses SFTP, but with a nice GUI, but again, I'm not writing my own programs here.

Categories, I put as keys are always fixed for me and I'm getting paths to them immediately without need to make find/grep each time

No offense intended, but you're still spending far more time than I'm talking about. Setting up unix servers with huge raid drives, finding an out-of-state site to stash it, setting up secure data transfers, devising your own method of assigning metadata to files or some kind of personal database file system....

I understand, for a geek, this isn't a rediculous expense of time, since it's also a hobby and a source of fun and entertainment. However, to grandma (and even me) it's just too much.

When I talk about making photos "easy to find", I'm talking "easy" like Apple's iPhoto is still a bit too complicated, in that you have to assign keywords and ratings manually, which many users aren't going to bother with after a certain number of photos.

When I talk about easy to access, I'm talking about the process being relatively transparent, i.e. easier than connecting to an FTP site. Like you wouldn't need to know that it's "not on your computer".

When I talk about affordable, I'm talking about something like $100 total, or a $10 a month service (for personal use).

In case I'm not being clear, I'm not asking, "What's a good, cheap backup solution, available today?" I'm saying, the state of data management technologies is not currently sufficient for our ever-expanding set of data. We need better search methods for all sorts of data (not just text). We need transparent backup and archival methods (transparent both in the backup and the restore). We need more than solutions for businesses who can employ a big staff and thousands in hardware, and more than solutions for geeks who can roll their own. We need solutions so that Joe Schmoe can take digital photos to his heart's content, can create a digital music library as large as he wants, and not need to worry about sorting through the data or losing it.

Re:DATA DATA DATA

[nine-times]

I think there's something to this idea, but I think it will ultimately require that files can be search on something other than "keywords".

Of course, many of these proposed filesystems allow for something like, "Give me all my jpg's that are larger than 640x480 and were created later than Jan 1st." So, already, we have more than keywords.

However, I still don't think it's sufficient. If I have thousands of photos, is it really reasonable to expect that I am going to be comprehensive about adding keywords to each? I mean, enough keywords for each photo that I can say, "Find that picture I took of the waterfall and a woman swimming underneath"? GIS can do this somewhat, but only because it's pulling metadata from the pages that link to the photo, and even then, it's not really reliable enough.

Our big hope, I think, is that it will be possible for pictures to be automatically analyzed for content. So, as a simple example, the computer might be able to tell the difference between a portrait and a landscape. Between a child and adult? A man and a woman? How far can we go with this?

Will computers be able to search music by whether it will get you pumped up or whether it will sooth you? Whether it *sounds* fast or slow? Will I be able to set my iTunes smart playlist to find me 50 "sad" songs out of my library?

I think this is the way things need to go, but it's certainly a "grand challenge" to get these sorts of capabilities working properly in consumer-level computers.

Re:DATA DATA DATA

[Fareq]

please forgive the obvious here:

You are correct. That is why it's not called Data Technology.

However, I think the key is that people want information and computers store only data. "Data Mining" is the science of extracting a small amount of information from a mountain of data. I guess it's a bit of a misnomer.

Gold Miners mine through a mountain of quartz looking for gold.

I don't know what kind of structures silver is in, but its the same deal, Silver miners are seeking silver.

The last thing Data Miners want to find is more Data. They want to extract the Information. After all, when was the last time you saw a Dirt Mine?

What are some other worthy computing challenges?

[IWantMoreSpamPlease]

How about an OS that doesn't suck?

I'm still waiting for things promised by Y2K

[bigtallmofo]

What about all the fanciful things we were supposed to have "By the year 2000!"?

What a joke that turned out to be. I'm still making calls with an audio-only phone and I have yet to come across a practical hover-car.

Re:I'm still waiting for things promised by Y2K

[eln]

Hover cars are redundant, since the way my wife drives, the tires rarely touch the road anyhow.

What about ...

[blogeasy]

a decent IT system that can manage the projects we've been waiting for. Namely, the flying car and Duke Nukem Forever. One day we'll see this future materialize.

Cell phones

[Reignking]

I think we need to develop cell phones that can cook, clean, and drive my car. For $25. Oh, and I guess they need to be able to send and receive phone calls.

Re:Cell phones

[TommydCat]

I think we need to develop cell phones that can cook, clean, and drive my car. For $25. Oh, and I guess they need to be able to send and receive phone calls.

Talk about being married to your cellphone...

Speaking of simulating life...

[rewt66]

Here's a challenge: A patient comes into a doctor's office with a bacterial infection. Worse, it's one of those antibiotic resistant bugs. What we need to be able to do is:
- sequence the bacteria's DNA right there in the doctor's office (this part isn't really an IT challenge)
- from the bacteria's genetics, determine which antibiotics (out of all known ones) can effectively kill it
- if none can effectively kill it, ship the DNA sequence information off to the CDC's supercomputers, and have them automatically develop a new antibiotic that will kill the bug.

I figure that this is a challenge for the next forty years, not just for the next twenty.

Most important goal...

[Anonymous Coward]

Come on people, we need to break the one million mark on the number of different text editors for unix based systems!

How about

[cubicledrone]

Keeping people employed for more than five weeks?

nonclassical methods

[k4_pacific]

From TFA: "Journeys in nonclassical computation: Classically, computation is viewed mathematically in terms of algorithms, but there are other ways to look at it. These include rethinking the rigid classification schemes computers use and turning to others based on family resemblance or on metaphor"

I know! I'll develop a new type of database that is indexed by the degree to which the primary key sounds either "woody" or "tinny" when spoken. I'll make millions!!

How about this?

[HotNeedleOfInquiry]

A computer system that will pass the Turing Test.

Re:How about this?

[kaustik]

From ALICE:

Human: Can you go get me some food? ALICE: Sorry my body isn't attached right now. I'm stuck inside this computer.

Simulated Sex should be our next challenge...

[stephanruby]

Simulated sex should be our next challenge, sex has already helped us, and will continue to help us, in pushing the limits of what's technologically possible.

Re:Simulated Sex should be our next challenge...

[Anonymous Coward]

Do you mean a simulation other than your left hand?

Right hand?

A Slashdot Dupe Checker

[OverlordQ]

Should be easy right? Never the less it has stumped slashdot editors for many many years.

Too bad they're impossible

[Wylfing]

Unfortunately, none of these aspirations will materialize. IT in the U.S. and Europe is going to stagnate for the next 10-15 years, because the RIAA and MPAA (and their European equivalents) will continue doing everything they can to bring technology back to 1996 levels; and patents on algorithms and business methods will confound any new technology ventures.

Re:Too bad they're impossible

[mmkkbb]

Oh come on, there's more to IT than personal computers.

Web applications

[fiannaFailMan]

The PDFs were getting a little slashdotted so I couldn't fully RTFA, but here's what I see as an exciting area: Getting the richness and usability of the desktop application in a web-based application. The metaphor of the submitted 'form' and requested 'page' is very limiting. Imagine using Word in such a way that you had to destroy and patiently reload the page every time you wanted to embolden a bit of text or reformat a paragraph. The reach of applications has taken a step forward with the web, but in terms of usability a giant step was taken back.

This is where technology like Macromedia Flex comes in. I've seen this stuff in action, and the process of creating complex applications is so easy it's unbelievable. A field of sortable and stretchable columns can be generated with about three lines of code, and the data that goes into it can come from any application server you like.

Sure, anything that uses the Flash player gets a hammering on Slashdot, but I sense that times are a changing around here and more people are starting to wake up to the potential of this stuff, even if it goes a little against the open source ethos of the place.

BTW, if you're a member of the "Flash sucks and I hate it because some people used to abuse it by making annoying animations with it" brigade, see my journal where I've already refuted your half-baked criticisms.

Re:Web applications

[NoMoreNicksLeft]

Too bad for the macromedia fanboys that the answer to your problem is called "SVG+XML+Xforms".

Re:Web applications

[fiannaFailMan]

It never ceases to amaze me that the anti Macromedia brigade so quickly rolls out their rebuttals along the lines of "Oh but if you do a week's worth of DHTML coding, use a few iFrames here and there, throw in a bit of server-side trickery, and anything else you can do in SVG, what could be simpler?" while casually ignoring that Flash or Flex can do all of this in a single easy-to-use package.

In any case SVG doesn't have half the abilities of Flash and it definitely doesn't have anywhere near the same level of browser penetration, hence the maturity of Flash. I remember someone telling me two years ago about how SVG was going to render Flash obsolete. Two years later and I have still yet to see a single SVG file rendered in my web browser, to say nothing of a job ad asking for this skill.

Here's a challenge...

[joshsnow]

What are some other worthy computing challenges?

Making Firefox on Linux as quick as Firefox on Windows... ;-)

Idea for Linguistic Intermediate Language

[Rie Beam]

How about an intermediate computer linguistic language for translations?

Let's say there's a chatroom with a guy from Poland, a girl from Japan, and a duck (this is not a serious example, obviously, and why they are in this chatroom is left to the user's imagination). The duck sends his message, and it gets scrambled into the intermediate language. This language can now be translated directly into any local dialect, without having to translate the message for each seperate language being used, or without the user having the know the language. Just imagine - a user from Russia chatting with a user from Mexico, and neither knowing the other is anything but their native tongue. Of course it's not meant to be a cultural mask or anything - certain language / cultural barriers would of course be present, but at least this is better than having to run to Babelfish every few seconds.

Re:Idea for Linguistic Intermediate Language

[Lord_Dweomer]

I wish they would make a Babelfish IRC pluggin that could automatically translate.

Verifying compiler? Correctness proving tools?

[SashaM]

Verifying compiler? Correctness proving tools? Two words - Halting Problem.

Re:Verifying compiler? Correctness proving tools?

[SashaM]

Maybe you'd care to expand on those two words to explain why you don't think that there are classes of computational processes for which classes of specification can be proven as met, or why you don't think this is useful...

There most certainly are such classes and classes, but the proving cannot be automated (except for non turing-complete languages). A computer can verify that a "proof" is indeed a proof, but it cannot produce such a proof itself.

Perhaps if every binary came along with a proof of its correctness, a verifying tool could check that the proof is correct. This would, however, just shift the burden to the developers, who would have to prove everything they write is correct. Maybe some language or tool could make proving correctness easier, but I don't see how it could make it significantly so, since, again, it could never be automated (and I take something that can't be automated to be difficult).

Re:Verifying compiler? Correctness proving tools?

[SashaM]

This is all well and nice, but the halting problem is just one small example of an undecideable problem. In fact, every nontrivial, semantic problem about computer programs is undecideable ("semantic" means that the answer only depends on what the program does as opposed to depending on the program itself. "nontrivial" means that the answer isn't the same for all queries).

This narrows the set of decideable problems to ones that are either:

• Non semantic - does this program compile? Does this program use recursion? etc. But then we've been solving such problems (automatically) for a long time.
• Trivial - these really aren't interesting problems.
• Not about computer programs - but we're talking about computer program verification here, so these programs may not be written in a Turing complete language. There are interesting non Turing complete languages (regular expressions for example), but they're not something you can write "real" programs in.

Basically, the point is that by restricting yourself to something verifiable you've restricted yourself way too much.

Re:Verifying compiler? Correctness proving tools?

[SashaM]

Oh, and I think you're a bit confused about Goedel, but then so am I, so I won't comment on it ;-)

Re:Verifying compiler? Correctness proving tools?

[gustav_mahler]

You're right. We can't write a program that takes any program as input and returns whether or not it halts in finite time.

We also can't write a program that verifies whether or not a given theorem in a sufficiently powerful formal system is true. This doesn't seem to stop us from doing math, or even writing theorem provers.

There are many, many programs about which we can programmatically verify generally undecidable properties. And if we can't for a given input, we can either disallow it as input, recognize it and give up, or just let our program loop merrily. In fact, I would be surprised if there is a real-world program out there now or in the future that can't written so that it can be be verified to halt or not. Has anybody been stopped from necessary computation by Post's Correspondence Problem? Doubtful.

There is a whole field of creating these types of compilers known as Proof Carrying Code. The idea is that a user specifies a security policy detailing what properties a program needs (halts, memory-safe), then a compiler automatically supplies and bundles with the code a proof if one exists, which is then verified before the program is run. This is real technology that works on large classes of real programs.

Re:Teleportation

[CosmeticLobotamy]

This distinction is important because we will learn to telecopy objects and telecopy live organisms before we learn to teleport them.

Helloooooo, lawsuit.
-.+AA

Rechargeable?

[Ironsides]

I thought fuel cells weren't rechargeable. As in rechargeable without pumping more Hydrogen into them. If it's not possible to recharge them as easily as you can a battery, it's not gonna succede very well. I don't think people will want to have to "Hydrogen up" their batteries like the "Gas up" their car.

Re:Rechargeable?

[plover]

You're correct in that they're not "rechargeable", they're "refillable."

Fuel will probably be available in cartridges that are shaped to fit the manufacturer's equipment. Replacing them will need to be as easy and fast as changing batteries. Don't forget that current fuel cells are designed with on-board cracking of methanol, which allows for liquid fuel rather than having a pressure tank of pure hydrogen. It will make things much more convenient, although at the possible expense of some size/weight, as well as lower energy density of the fuel.

Probably the biggest drawback will be that each manufacturer will likely specify somewhat different shapes with patented, incompatible fittings in order to "maximize brand loyalty" (lock you in to their refills.) As far as I can see, making it inconvenient would be the quickest way to kill off adoption, but manufacturers usually see things differently than I do.

Solve the spam problem

[menscher]

Seriously. It seems like it shouldn't be that hard, but it is. So let's solve it already!

Biggest Problem in that Scenario

[Inhibit]

You forgot

- Get the patient to take the antibiotic all the way through

That's the crucial missing step that's let the nasty bugs get this far :).

Re:Biggest Problem in that Scenario

[Daniel Dvorkin]

Here's a very simplified explanation:

The problem is that resistance isn't either/or -- that is, it's not as simple as saying a particular strain of bacteria is resistant or it's not. All strains have greater or lesser degrees of resistance; more precisely, individual bacteria within the population have greater or lesser degrees. When you're on antibiotics, the bacteria tend to die off in, pretty much, an exponential decay curve. Once the curve drops below a certain level, the remaining bacterial population is insufficient to maintain the infection; your immune system is fighting the infection too, of course, and it can take care of the remaining bacteria, which are the more resistant ones, one the less resistant ones are killed off by the antibiotics.

So what happens when you stop taking the course of antibiotics halfway through? Well, where you previously had a bacterial population consisting of some bacteria with weak resistance, some with moderate resistance, and some with strong resistance, now you only have the latter two categories. And these are going to continue breeding, and your immune system is going to spend its resources fighting them equally, without preference as to which is more or less antibiotic-resistant -- which means more of the bacteria with greater resistance will survive and grow. OTOH, if you'd finished the antibiotics, only the most resistant bacteria would be left, and your immune system could probably finish them off on its own.

To top it off, resistance requires an expenditure of energy on the part of the bacteria -- you're quite right that many such critters have non-expressed resistance genes already in their genomes; the reason these genes aren't usually expressed is because doing so takes energy the bacteria would usually prefer to devote to feeding and reproducing. So in a patient who doesn't take antibiotics at all, the percentage of resistant individual bacteria is going to be very low. This means that taking half a course on antibiotics is the worst possible course of action: if you take the whole thing, you'll probably end up killing off the entire infection; if you take no antibiotics, you'll either get better or you won't, but either way you won't encourage the formation of a resistant strain.

And the reason that shorter courses of antibiotics are being prescribed is that, quite simply, many newer antibiotics work more quickly. That's the only reason. It has nothing to do with some magical discovery that the traditional ten-day course was longer than it needed to be.

The worthy challenge

[reshin]

Save the environment; most other things can be delayed. Discover efficient alternative energy sources to plant and fossil fuels; develop the materials and processes to implement these alternatives; build more detailed environmental models to aid in the study of the effects of pollution and the effects of tearing down natural habitats.

Brute force AI timeline

[Anonymous Coward]

We know to some extent how the human brain functions, at least at the level of neurons and synapses. A sufficiently accurate simulation of 10^11 neurons and 10^14-10^15 synapses should produce a human level intelligence by brute force. Clever AI software design may require less than this, but I claim that it is an upper limit.

The exact computation required to simulate a neuron sufficiently accurately is not known exactly, but we can put some reasonable estimates to it. I use 1 synapse firing = 1 bit +- factor of 30, which leads to a human equivalent = 3,000 Tflops (range 100-100,000 TFlops).

I will take as a proxy for 'largest computer available for AI research' the 500th computer listed in the top500.org list of most powerful supercomputers.

The trend has been for the #500 machine to grow at 93% per year in performance. A factor of 30 uncertainty in required performance thus only leads to a 5 year unceratinty in date.

3000 Tflops for the #500 machine would occur in 2017 at historical trend rates, to which I would add 5 years for software development/AI training, so the 'danger zone' for superhuman AI starts at 2017-2027.

SETI@home runs 65 Tflops currently on a distributed network, which is barely below my low end 100 Tflops estimate, so the risk of a runaway intelligence on a distributed network is non-zero (whether malicious or well meaning). The risk from a top ranking supercomputer is lower in my opinion. The #1 machine clocks 70 GFlops, but the top ranking machines are operated in a much more controlled environment.

If I was asked what will seal our doom, I would say it's the playstation 3. It will contain a 'Cell' processor jointly developed by IBM, Sony, and Toshiba. It's designed to be highly parallel, and it will be produced in mass quantities which will make it cheap. Thus it will will be well suited to MPP type supercomputers. I for one welcome our new Sony-based overlords...

Re:Brute force AI timeline

[divisionbyzero]

Of course, all of this assumes that human intelligence can be simulated by computation, in the classical sense. By simulation, I mean that a machine would demonstrate human-like intelligence. I don't think this is the case, but I don't see why we shouldn't pursue the brute force strategy, at least to rule it out. I don't really buy your guess on the number flops that would be necessary either, even if I assume that computation could simulate human-like intelligence. Every few years the number gets bumped up by an order of magnitude. If you look ten years ago, they were saying all we would need is 3 Tflops. Obviously, that's not true unless the problem is software-based. The real fundamental problem is that we do not even understand what intelligence is. It's hard to simulate something that you don't understand.

As for thought itself, I seriously doubt it works in the same way that a hardware simulation that you are describing would work. Think about how much energy would be required and how much heat would be generated compared to a human brain. Biology simply doesn't work in that way. Look at protein folding. It's extremely computationally intensive to determine the way a protein will fold, but biologically the process of folding is relatively simple. It's the same with thought. If we could figure out how the brain works, then we could probably simulate it with hardware that we could make now.

Re:Brute force AI timeline

[TimothyTimothyTimoth]

A deeper brute force simulation at the atomic level would be just a few more Tflops/years away. The problem of understanding at a "higher level" the thing that you are atomically scanning then modelling with basic physics is moot, especially if the full brain/body is simulated.

(That is, a good enough atomic-level brain/body simulation would still respond "don't remind me" when asked about it's last birthday, just like the human being being simulated.)

Whether anybody was home would be one for the philosophers, but such a simulation, of say a computer researcher, could work, and earn money just as well as it's original. So capitalism would pursue it. And it will rise in speed with hardware advances (which will increase correspondingly). So FOOM!

Re:Brute force AI timeline

[Boronx]

Obviously, that's not true unless the problem is software-based.

I think it *is* software based. We don't really all of the mechanisms that cause synapse formation or alteration, and I saw some research last year that suggested that synapses and neurons may not be the entire answer to the brains computational power. There were some cells thought to be support cells that have shown indications of communication with eachother and neurons.

When someone says brute force, I take it to mean a simulation at near molecular level, so that we don't necessarily have to know *how* it works, as long as we have all of the right components, it will just work. Careful tracing of the operation would then lead to insights about how real brains work. This, of course, would require a great deal of computing power.

Re:Brute force AI timeline

[Finuvir]

Intelligence, true intelligence, may not require consciousness. We don't know. Consciousness, qualia, the feeling of awareness is the aspect of mind we know least about (and most about in another way I suppose). A human-like intelligence may well be harder to achieve than another sort of intelligence, however you might decide on that. But when you get right down to it, we know that machines can be made that have human intelligence. They're called humans. Unless we resort to superstition to "explain" our intelligence and awareness it's clear that AI is in principle possible.

Re:Brute force AI timeline

[divisionbyzero]

You might be referring to glial cells, but, anyhow, computation is computation. Thought *may* be something that cannot be simulated by computation. See the problem with most computation is that it goes from one determinate state to another. It's somewhat like a deductive system. One deduction follows from another inevitably. Humans don't think that way. Clearly some forms of thought can already be simulated to a certain degree, logic, mathematics, and even to a certain extent science. For the most part these are very much based on rules and do not require what I would consider "creative" thought. In other words, developing the axioms from which deductions would be made, developing new kinds of maths, designing new experiments, or coming up with it's own rules.

Incidentally, the robot that can do science is the most interesting to me because it seems to be the most creative. It even comes up with new experiments. http://www.trnmag.com/Stories/2004/012804/Robot_au tomates_science_012804.html
Yet it has no idea what it is doing and it is only following rules. It can't create it's own rules. This example is the one that seems to indicate the immediate future of AI to me.

Of course, you never know. It could be that thinking can be simulated perfectly fine with computation, but I don't think it is an obvious conclusion, nor are the similarities between the brain and a computer that great. I definitely think it is worth researching, but I just don't think it will end up turning out that way. But that's just a hunch. I don't think there is definitive evidence either way because, like I said, we don't know how the brain works.

A simulated organism?

[ailwardraeg]

Imagine the possibilities of such an invention. Testing chemicals and medicines on animals would become an expensive, backwards way of ensuring the safety of consumers. Perhaps the simulation of an entire virtual organism would not even be necessary in many cases.. only the molecules (and many properties thereof) that make up the portion of skin and flesh to be tested against topical agents, for example. It sounds as if in the end it would have to be a sort of mini-Matrix.. maybe a virtual area 2 meteres squared where the global constants of Earth gravity, Newton's laws, etc. are emulated. This is beginning to sound like it would require a unified theory of everything. Perhaps some clever people with enough money to research this will figure it out.

It would most likely require quantum computers to have become a reality, so let's hope those come around in the next ten years. (Die, x86! Die!)

Re:Simulated Sex

[Cid Highwind]

Mod the parent funny if you want, but sim sex would drive debelopment of lots of cool new technology. The requirements are mind-boggling.

First, before any code could be written, you would have to integrate biology and psychology into a single unified theory just to get a handle on what sex really *is*.

Second, you would need code and hardware capable of simulating a human mind and body. Even the NSA's "It doesn't really exist, we promise" crypto-crunching supercomputers would choke on that task.

Third, you would need an interface. A full model person is going to be impractically large and heavy(*). It would also be difficult to change after it's built (and I don't think many potential sim-sex customers are going to want sim-monogamy). The best solution would be a direct neural interface, but that would require more new technologies.

If somene had the motivation (and the knowledge, and the money) to make sim sex work, it would be a huge boost to all sorts of science and technology. Get busy, pornographers!

* Don't bother posting the obvious joke about how most /. readers (and their partners) are already impractically large and heavy. I'm sure everyone reading has already thought of that one...

They need to learn basic compsci

[DunbarTheInept]

To achieve the goal of building dependable computer systems, the scientists suggest building a verifying compiler, a tool that proves automatically that a program is correct before allowing it to run -- something first written about in the 1950s.

This, admittedly was in the summary text in the magazine, not the article by the scientists themselves, so it could be a case of "idiot summarizing it wrong", but there just is NO WAY to do what they are talking about. No how, no way.

To prove a program correct requires that you run it in a test environment. If you run it, and it is not correct, you get the same problem in your test run that occurs in the real run. Therefore you cannot test for a program's correctness automatically in a compiler. For example, any program trying to detect if a loop is infinite will itself end up looping infinitely when it encounters one and tries to check it.

Re:They need to learn basic compsci

[Animats]

just is NO WAY to do what they are talking about.

Wrong. Machine-verified proof of correctness is quite feasible. We did it twenty years ago. The DEC SRL people did a nice proof of correctness system for Java in the 1990s, before Carly shut down DEC research. It's hard to build such systems, but not impossible. The theory is well understood now, which wasn't true when we did it.

It's not that hard to prove loop termination. You must define some measure which, for each iteration of the loop, decreases. For many loops this is trivial. For most loops it isn't too hard. For loops so complicated that it's hard, add a loop counter to detect non-termination as an error.

Proof of correctness went out of favor because C won the programming language battle, and C semantics are so ill-defined that formalization is hopeless. Java, though, isn't bad.

The "design by contract" people have the right idea, but it's hard to retrofit design by contract to C++ in a sound fashion. If you're going to have object invariants, you need to insure that control never enters the object when the object is not in its stable state. This is a constant problem in C++, because you can call out of an object and then back in. (GUI systems are notorious for this.) You need to be explicit about inside/outside issues. There needs to be an explict way to say "control is now leaving this object" at the point you call something that could call you back. Without that, object invariants are meaningless.

Hardware proof of correctness tools are widely used. Look up VHDL verifiers.

Re:How about an OS as good as VM/370 with a GUI?

[saden1]

Solve one, jus one, NP-Complete Problem.

A controlled bomb

[DigiShaman]

What you're advocating is a controlled "bomb". No seriously, do the math and you will find the energy density is rather tight.

BTW, I used to work at Dell in the safety division. I've seen video footage of laptop batteries explode. Now mind you, these were 3rd party batteries people of bought off e-bay and the like. So technically, these issues are NOT cause by Dell. None the less, it is something to be concerned about.

Re:Actually...

[FleaPlus]

I'm not quiet sure you understand. With a verifying compiler, the programmer defines what is considered correct. The compiler verifies that the program is correct, according to the definition.

Of course, if the definitions are wrong, all bets are off, but it's still an incredibly useful thing to have.

For another example, think of software test suites. Nowadays, you have a programmer explicitly defines a score of situations and checks to make sure that these situations fit defined requirements. With a verifying compiler the programmer still has to define the requirements, but the computer is also able to mathematically prove that any possible set of inputs and situations will obey the defined requirements.