WA: What citizens need to know about the AI Task Force
SB 5838 was passed out of committee on Monday and is now on the calendar to receive a floor vote in the House. If you’ve been following along this session, you know this is a bill we are deeply concerned about for a number of reasons. I (Julie Barrett) engaged with citizen activist and friend, Teo Morca in an email exchange following the passage out of committee with amendments. To be honest, the technicalities of AI is a little above my paygrade and it’s so helpful to have someone like Teo to advise and teach me and our team. His response was so well-articulated, I asked if I could share it publicly and he obliged.
By Teo Morca
This AI task force is inevitable and there is no way around it. However, AI and especially generative AI (because it is so new, powerful, and the most accessible to all at the moment) IS an arms race. We do not want the deck of AI-generated policy stacked against us by the radical progressives developing a monopoly on its use at the state and local levels. This is EXACTLY what Inslee and Ferguson and the more tech savvy radicals in the legislature want: to get a framework-wide lock on governing the use of AI to form policy and to anticipate more effective moves and countermoves to fight Republicans.
Without a strictly guaranteed and enforced parity of political and ideological perspectives on these task force committees, this is simply the Dems stacking the deck without anyone challenging them.
To levelset understanding, Artificial Intelligence is the broader concept of enabling a machine to act like a human; Machine Learning is an application of AI that allows machines to extract knowledge from data and learn from it autonomously.
The lawmakers hand-wringing about algorithmic bias do not understand algorithmic bias. True algorithmic bias is when the machine learning model or "algorithm" makes a mistake, and that mistake results in a misleading or unfair outcome directly influenced by that bias. HOWEVER, just because an output becomes "unfair" does not mean it is wrong or inaccurate. Algorithms learn and improve over time as they are applied to more and larger sets of data. When the data show patterns, the algorithm - just like a human - learns and adapts to those patterns.
Because of this, and because the algorithms improve with more data, we MUST be prepared for AI to become biased over time based on new learnings and experiences, just as humans do. For example, if 4 out of 5 engagements with a penguin result in the penguin attacking you, and only 1 out of 5 engagements with a bear result in the bear attacking you, you will naturally develop bias concerning bears and penguins. This is still a bias but it is not inaccurate, or something that needs to be somehow modified to correct. The reality simply is the reality whether it was the outcome you hoped for. This is how machines and humans learn.
There’s a lot of misunderstanding of bias. Bias is not inherently bad when it ensures that we don’t have to re-learn not to touch a hot stove every day based on experience.
When we begin to modify algorithms that are already well-established with new coefficients or weights to manipulate outputs to express a desired outcome, we will break things. I am deeply concerned about the task force breaking things to force desired social and racial narratives.
The problem with this SB 5838 is that it assumes that all outcomes that present in fine-tuned or trained models as somehow "unfair" to a particular demographic are somehow "wrong" and must be corrected or overcome through constant vigilance by a politically left-biased task force. What if the model begin recommending that police begin policing a specific neighborhood block on Rainier Avenue more intensively than they police Island Crest Way on Mercer Island? You know full well that the Dems are going to cry bias and racism and try to manipulate the machine learned policing algorithm, when the TRUTH is that - just like a human - the machine has learned that if more crime happens on Rainier Avenue, you police it more than Mercer Island. This is not "bad" bias or even algorithmic bias even if the result is that more people of color get arrested; it simply the result of an algorithm that has learned from experience.
So, in my opinion, AI governance must NEVER be solely in the hands of a politically biased agency, which the AG's office and all of its agencies are. This is how true bad bias gets integrated at training, making it impossible for output fine-tuned models to be unbiased.
This bill expresses concern about AI used to violate the civil rights of certain communities, but this is disingenuous. Our AG has demonstrated comfort with violating rights to support an agenda. This task force may be used to safeguard certain civil rights but decimate others, depending on the target community.
The AG’s office has conflicts of interest and biases that will compromise the fairness, transparency, and accountability of AI governance. For example, the AG may direct the task force to enable its own purposes, such as criminal investigations, prosecutions, litigation, or drafting laws that infringe or impair constitutional rights, which raise ethical and legal issues. The AG's office also favors certain stakeholders, industries, and agendas over others, which IS bias, and which will undermine the diversity, inclusivity, and representativeness of AI governance for all stakeholders.
An amendment requiring any task force or committee governing AI to guarantee equal political and ideological representation and ensure the same number of contributors or members from each political party and perspective, REGARDLESS of political majority or minority is needed.
A group made up of apparently diverse people who all believe the same thing is not meaningfully diverse, this is an echo chamber.
This chart illustrates how bias is integrated into technology and machine learning algorithms. If all communities that are informing the research, development and implementation of machine learning and AI are themselves biased, the output model(s) will be biased.
Without a guarantee in statute that the task force and all committees under it will have exactly equal numbers of members from left and right, Democrat and Republican, Progressive/Liberal and Conservative, etc. viewpoints REGARDLESS of majority/minority in legislature, then the danger of this exercise being used to stack the deck is assured.
SB 5838 passed the House on 2/29/24 with strong bipartisan support
Please consider contributing to Conservative Ladies by making a donation or upgrading to a paid subscription/membership to support our work to defend and advance conservative policy in Washington and beyond. Thank you!