AI Ethics & Policy Insights
The AI ethics & policy info we should know - simplified. Week of November 6 , 2023. Issue 15.
Major news this week:
OPENAI ANNOUNCEMENTS
OpenAI hosted their first in-person event this week!
On the first day of their event they announced that GPT-4 Turbo is here! It’s faster, less expensive for developers, and it allows people to create their own custom versions of the GPT Chatbot.
It’s not altogether revolutionary, but it’s pretty impressive, including its ability to take in up to 300 pages of context.
META & POLITICAL ADVERTIZERS
Meta is banning political advertisers from using generative AI tools for creating political advertisements, according to Reuters. The goal is to limit the spread of misinformation. While this limits access to their tools, this policy doesn’t appear to limit political advertisers from posting AI generated content created with other tools.
Follow up on the UK AI Safety Summit
The Context:
The world’s first AI Safety Summit was hosted by UK Prime Minister Rishi Sunak last week.
The summit was hosted at the UK’s famous Bletchley Park, well known for being the home to Britain’s world war two codebreakers, including Alan Turing. The choice of this location was meant to symbolize advances in computing through AI, AI’s promise in defending against existential threats, and AI’s promise for the benefit of the world.
The goal of the summit was not to agree on policy, but to generate dialogue and spur movement toward collaborative actions around AI safety.
Attendees included delegates from 27 governments around the world as well as global business leaders from companies creating AI. Sam Altman and Elon Musk, who famously split as collaborators on OpenAI over differences in opinion about whether the company should be for-profit, were among the business leaders in attendance.
Wins for the summit:
SUMMIT TONE OF COLLABORATION & OPTIMISM. Some felt that the US AI Executive Order overshadowed the UK’s AI Safety Summit, but James Cleverley, the British foreign secretary stated, “This isn’t about hoarding, this is about sharing…this is something we want everyone involved in. It’s not exclusive, it’s inclusive.”
Max Tegmark, an MIT professor behind the letter asking for a pause on advanced AI LLM creation said that the summit was an encouraging success stating that it, “has actually made me more optimistic. It really has superseded my expectations…I’ve been working for about 10 years, hoping that one day there would be an international summit on AI safety. Seeing it happen with my own eyes - and done so surprisingly well - was very moving.”SAFETY TESTING. AI labs present (Open AI, Google DeepMind, Anthropic, Microsoft, Amazon and Meta) agreed to give the UK “early access” to their models for safety testing, although details are sparse on what “early access” means.
UK AI SAFETY INSTITUTE. Sunak announced that the UK Frontier AI Taskforce - originally created to conduct AI safety evaluations - will become permanent as the UK AI Safety Institute. According to Sunak, “Until now, the only people testing the safety of new AI models have been the very companies developing it. That must change.”
BLETCHLEY DECLARATION ON AI. The declaration was signed by 28 states, including the EU, and establishes that AI poses risks in the short and long term. By signing the declaration, each state represented agreed to collaborate on identifying and mitigating risks.
From the declaration: “We recognise that this is therefore a unique moment to act and affirm the need for the safe development of AI and for the transformative opportunities of AI to be used for good and for all, in an inclusive manner in our countries and globally…alongside these opportunities, AI also poses significant risks, including in those domains of daily life.”
HIGH LEVEL FRAMEWORK. At the summit, a high level framework around types of AI risks was established. It’s in its nascent stages, but is a good first step toward a global standard.
CHINA IN THE HOUSE. China was invited and included in the Safety Summit. While many, including Liz Truss, former Prime Minister, urged that China’s invitation to the Summit be rescinded, having China participate in this summit, even symbolically, is a win.
STATE OF AI SCIENCE REPORT. It was determined that an output of the summit would be an annual “State of AI Science” report. The inaugural report will be chaired by Yoshua Bengio, one of the “godfathers of AI”, and winner of the AMC Turning award. The group writing the report will including academics in AI and will have an advisory panel drawn from countries that attended the summit.
Weak points for the summit:
LACK OF CIVIL SOCIETY INCLUSIVITY. Vidushi Marda, a delegate from Bangalore, India, pointed out that the conference mostly included US and UK civil society. “Most of the consequential decisions are pretty opaque to us, even though we are in the room.” More than 100 organizations and individuals signed a letter criticizing the sidelining of civil society from the Summit and called for increased openness.
PROTESTORS. A group of protesters from the group Pause AI protested at the summit, sharing that polling indicates people would like to see a pause on AI development. This movement comes on the heels of the open letter signed by many AI professionals urging that development of systems more powerful than OpenAI’s GPT-4 be paused.
WHO DETERMINES REGULATION. AI Now Institute executive director Amba Kak indicated that there are concerns about industry guiding regulation, rather than regulation guiding industry. “The context to all of this is that we’re seeing a further concentration of power in the tech industry and, within that, a handful of actors. And if we let industry set the tone on AI policy, it’s not enough to say we want regulation - because we’re going to see regulation that further entrenches industry interests.”
THE LUXURY OF LONG TERM HARMS. Conference attendees also indicated that there seems to be a disproportionate focus on the longer term harms that could accompany AI development and growth. The focus on those far-off theoretical harms takes away attention from the harms that are happening right now. While some praised the Summit for focusing on a range of AI threats, others felt the conference was too heavily leaning toward existential or “frontier AI” threats. Professor and director of Research at the Oxford Internet Institute Brent Mittelstaeds stated, “Frontier AI should not be used as an excuse to avoid regulating the well- established harms of today’s AI systems.”
OPEN SOURCE AND AI. The debate for open-sourcing AI models for the purpose of safety research is ongoing and challenging. “In the discussions I was in, there was quite a bit of agreement that we needed to go beyond a binary open/closed-source debate, and look at more nuanced solutions, but still differing perspectives on what that needed to look like and what should be prioritized.” stated head of AI policy for the Centre for Long-Term Resilience Jess Whittlestone.
Yann LeCun, Vice President and Chief AI Scientist at Meta stated on X that he signed a letter to President Biden urging that any regulation should promote open source AI platforms, not hinder them.
Where we go from here:
The next AI Safety Summit is planned in South Korea in 2024.
Sources: TIME, GOV.UK, The Guardian
AI Policy Updates
The EU AI Act
The EU AI Act is expected to be finalized by the end of 2023. It is currently going through another round of the “trialogue” which is a dialogue between three people or groups. In this case the trialogue takes place with the European Commission, The Council of the European Union, and the European Parliament.
The EU AI Act is expected to be passed by June 2024. Following its passing, companies and organizations conducting AI-related business in the EU will have a two year grace period to get aligned to the requirements.
TikTok of the Week
AI Executive Order: PRIMER II
Tiktok failed to load.Enable 3rd party cookies or use another browser
Ask your AI Question.
Submit your AI question here - it can be a responsible AI, AI policy or general AI question. I’ll pick one to respond to every week!
This week’s question: What groundbreaking work are we using AI for in the medical field, and how can we prevent bias in this field?
(Submitted by: Nicholas deLaurentis)
This is part II to the question listed above. Last week, I shared my response to the first part and this week I’ll address the second part: How can we prevent bias in medical field uses of AI?
My response: Bias in general is a significant concern in the medical field. People are concerned that their gender, race, socioeconomic status, or other similar factors will determine what kind of care they receive. If a doctor has bias against those who use drugs, they might administer care less diligently. This same scenario can occur with some details changed, and it amounts to real harm for real people, often at a time when they feel most vulnerable.
The reason AI use in the medical field is so top-of-mind is because this technology - intentionally created to mimic humans - can mimic the bias, too. It’s a legitimate concern. So how can we prevent bias from making its way into AI technology in the medical field? There are two ways, as I see it, and they depend on humans.
1. We can choose not to use AI in scenarios where standard of care could be compromised by bias. Some doctors use speech-to-text AI to take dictation while a doctor examines a patient, and then they have AI consolidate the notes into a summary. This can be done without bias-infused technology. No judgement or diagnosis is given. The AI simply ingests the information and summarizes it. It’s very limited in functionality and isn’t informed by external data. Doctors and tech decision makers can choose to use AI that helps, but that doesn’t have capacity for bias.
2. We can have governance boards and check points for AI technology that could be bias containing, or which has outputs that could exacerbate human bias. If, hypothetically, a medical institution chose to use facial recognition technology AI to try and identify people seeking pain medication for illicit purposes, this is an example of where bias can show up at several points. There can be bias in the facial recognition system itself. There can be bias in the data it was trained on, and, adopting the garbage-in, garbage-our principle, the bias that goes in, would produce a biased output. From there, we could see bias in medical professionals who are scanning to see if there’s a match in the system for those who regularly seek medication from hospitals. Whenever there is AI that could be biased or exacerbate human bias, there should be processes, systems, checks and balances and the like in place. There should never be technology making decisions or judgements without several diverse perspectives evaluating the output and approving the decision or judgement.
Bias will exist in the medical field because it exists in humans and it can exist in AI technology. It’s up to the leadership of hospitals and medical institutions to establish teams to evaluate incoming AI technology, set up processes and governance boards, establish standards and provide training for all personnel.
Let’s Connect.
Connect with me on LinkedIn.
Follow on TikTok, where I talk about developments in responsible AI.
Email me thepaigelord@gmail.com if you want to connect 1:1.