Slashdot Mirror
A New Bill Would Force Companies To Check Their Algorithms For Bias (theverge.com)
An anonymous reader quotes a report from The Verge: U.S. lawmakers have introduced a bill that would require large companies to audit machine learning-powered systems -- like facial recognition or ad targeting algorithms -- for bias. The Algorithmic Accountability Act is sponsored by Senators Cory Booker (D-NJ) and Ron Wyden (D-OR), with a House equivalent sponsored by Rep. Yvette Clarke (D-NY). If passed, it would ask the Federal Trade Commission to create rules for evaluating "highly sensitive" automated systems. Companies would have to assess whether the algorithms powering these tools are biased or discriminatory, as well as whether they pose a privacy or security risk to consumers.
The Algorithmic Accountability Act is aimed at major companies with access to large amounts of information. It would apply to companies that make over $50 million per year, hold information on at least 1 million people or devices, or primarily act as data brokers that buy and sell consumer data. These companies would have to evaluate a broad range of algorithms -- including anything that affects consumers' legal rights, attempts to predict and analyze their behavior, involves large amounts of sensitive data, or "systematically monitors a large, publicly accessible physical place." That would theoretically cover a huge swath of the tech economy, and if a report turns up major risks of discrimination, privacy problems, or other issues, the company is supposed to address them within a timely manner.
The Algorithmic Accountability Act is aimed at major companies with access to large amounts of information. It would apply to companies that make over $50 million per year, hold information on at least 1 million people or devices, or primarily act as data brokers that buy and sell consumer data. These companies would have to evaluate a broad range of algorithms -- including anything that affects consumers' legal rights, attempts to predict and analyze their behavior, involves large amounts of sensitive data, or "systematically monitors a large, publicly accessible physical place." That would theoretically cover a huge swath of the tech economy, and if a report turns up major risks of discrimination, privacy problems, or other issues, the company is supposed to address them within a timely manner.
When the facts say that men are on average stronger and taller than women, are the facts wrong? Some things are fact. Other things are uncomfortable facts. When facts conflict with beliefs (especially politically), which do you think will win?
Same for almost anything. Skin colour rarely matters, and given enough more direct data on factors that do matter, skin colour will have no predictive value, so the algorithm will ignore it.
If that were true then it wouldn't matter. However, either by nature or nurture, color matters. If it didn't Asians wouldn't be given penalties and other groups bonuses for college admissions. If we want to argue that race isn't important, and I don't think it is important, then we have to do away with the diversity quest and let it play out.
Bias is easily defined: Anything that doesn't match the subjective opinion of the government of the day.
These days that means that a hiring algorithm had better hire better than 50% women, and every ethnicity in relation to it's percentage representation in the population. What the algorithm is not allowed to do is take in to account any factors that might skew that, such as the applicant pool being predominantly one group, or the qualifications of individual applicants.
How about a dating site that decided you had a bias in favour of tall partners, or light skinned partners? There is no simple answer to this.
And how about a dating site that figures out you had a bias for a certain gender ?
Which has nothing to do with bias. Bias, in this context, is unwarranted assumptions. Men are on average stronger and taller than women, but a system which, say, ranks potential firefighter applicants using their gender as a factor instead of looking at their performance in the actual job is biased.
Sorry, but you haven't been listening to the left if you think the test for bias is about assumptions. Equal outcomes is very strongly being pushed as the measure for bias.
Do you have more males than females going into trades? That must mean a bias against females exists in the trades.
Do you have more Asians getting into STEM? That must mean a bias in favor of Asians in STEM.
Language is being redefined and weaponized to push people's agendas(I know, it always has). Today we have the definitions of equality, racism, bias, violence, assault and others being changed to better fit agendas. Racism being the grossest example because of it's importance and power. I grew up understanding racism to be discrimination based on race. Today though the push is on to redefine racism to be a combination of discrimination AND power. This turning the convenient trick then that 'whites' have all the power, so now only they can be racist, by definition.