Slashdot Mirror
Intel Predicts a $7 Trillion Self-Driving Future Where Over a Million Lives Will Be Part of the 'Passenger Economy' (theverge.com)
Intel has released a new study that predicts a $7 trillion annual revenue stream from the emerging passenger economy. In the report, Intel says that the companies that don't prepare for self-driving risk failure or extinction. Additionally, the report finds that over a half a million lives could be saved by self-driving cars over just one decade. The Verge reports: The study, prepared by Strategy Analytics, predicts autonomous vehicles will create a massive economic opportunity that will scale from $800 billion in 2035 (the base year of the study) to $7 trillion by 2050. An estimated 585,000 lives could be saved due to autonomous vehicles between 2035 and 2045, the study predicts. This âoepassenger economy,â as Intel is calling it, includes the value of the products and services derived from fully autonomous vehicles as well as indirect savings such as time. Autonomous technology will drive change across a range of industries, the study predicts, the first green shoots of which will appear in the business-to-business sector. These autonomous vehicles will first appear in developed markets and will reinvent the package delivery and long-haul transportation sectors, says Strategy Analytics president Harvey Cohen, who co-authored the study. This will relieve driver shortages, a chronic problem in the industry, and account for two-thirds of initial projected revenues. One of the bolder predictions is that public transportation as we know it today â" trains, subways, light rails, and buses -- will be supplanted, or at least radically changed, by the rise of on-demand autonomous vehicle fleets. The study argues that people will flock to suburbs as population density rises in city centers, pushing commute times higher and âoeoutstripping the ability of public transport infrastructure to fully meet consumer mobility needs.â
What does a computer playing a game with a strict set of rules have to do with self-driving cars? Nothing.
The rules of the game are simple and strict, but evaluation of the board is extremely hard and fuzzy. Comparable to driving, actually. There is a strict number of rules driving a car, but complicated evaluation.
Already we are seeing that processor speed is only marginally improving year over year.
My point exactly. The sudden breakthrough of AlphaGo had little to do with slowly improving processor speeds, but mostly with some guys coming up with a few clever ideas. And this year's version was running on dedicated hardware, not general purpose processors, so it could gain orders of magnitude improvement in a single year.
Wrong. In fact, massively so.
The whole reason that Go is so signifficant as a milestone for AI is because despite the ruleset being simple and straightforward, the amount of possible board configurations exceeds the number of atoms in the known universe.
Ask a good chess player why he made a certain move and he'll be able to give you a very well defined answer. "I moved the knight there because by doing so I have a guaranteed mate in 5 moves" or some such. Now go to a pro Go player and ask the same thing and they won't give you the same answer, they'll often give some variety of 'it felt right'. You cannot number crunch Go in the same fashion you can many other games simply due to the fact that the complexity and amount of possible plays far exceeds what even the current top of the line algorithms, let alone a human brain can handle. Thus there's a huge component of essentially intuition involved in Go.
This is why even a couple years ago people were still saying it's not possible for computers to beat humans at Go because it requires actually learning to evaluate moves entirely differently from chess. It requires a level of essentially creative thinking.
Most people do not think the progress will continue indefinitely but it doesn't have to do so in order for for computers to a) achieve human level general intelligence or b) achieve consciousness.
What he was trying to point out is precisely that these days the key developments are happening in the software, not the hardware.
That's the whole reason alphaGo is a good example. It's not that google beat Go because they suddenly got a slightly faster supercomputer able top crunch even more numbers, it's that they beat it by creating better algorithms that ran on existing hardware and learned to play the game better than humans.
Self-driving cars are the exact same thing. You're trying to teach a computerized system to play a game (traffic) which has fairly simple rules that even relatively stupid people manage to follow but has an almost infinite amount of possible outcomes (comparable to the amount of board positions in Go). the problems that remain to be solved are not related to processor power. We already have more than enough processing power in self-driving cars to vastly exceed the capacity of any human. The challenges to be solved have to do with machine learning and actually teaching that cars to become good drivers in various different weather and road conditions.
And just like with Go the car will also have to know how to deal with entirely new/unforeseen situations and pick the best move with no prior experience of such a situation.
That's why it's relevant. The key to AI is first and foremost the algorithms used, and those algorithms are improving at a pace which vastly exceeds the linear pace of hardware improvement, as evidenced by alphaGo.
"It is the business of the future to be dangerous" -Alfred North Whitehead