AI Safety

Recently, I have noticed that students have been using AI for homework help more and more, reducing the opportunity for thinking hard about a problem. I have also been reading reports of vulnerable children being victims of AI chatbots, or lonely adolescents forming inappropriate relationships with them.

As a STEM education provider, I believe that we have the best resources in order to understand what is happening in AI, and therefore the responsibility to use our knowledge to protect children and vulnerable individuals around the world through this period of rapid development.

The Problem

I do think that a lot of vulnerable young women, especially those from marginalised communities, are often left out of these conversations, and end up using AI in a way that could be detrimental to their mental and emotional health.

There are a lot of different things that we can do to help people through this time. Our starting point is education. Without education that is reliable and legitimate, individuals are more prone to trust misinformation that is being widespread, and feel lost in the discussions that occur in the community.

How We Can Help

I believe that communities that are marginalised from the frontier discussions in AI development need to be armed with AI safety awareness, in order to navigate and protect their futures as well as the values their communities are built upon, from poorly regulated AI technology proliferation. I also believe that more communities need to have a say in how AI technologies are designed, trained and developed.

Our initial solution to this would be to run education webinars for charities that serve students and individuals typically left out of these conversations, and develop a set of safety principles for councils or other policymakers to refer to. We hope to educate individuals about the dangers of AI and the various global factors that influence AI development, in hopes of equipping young people with more knowledge as they navigate their future in the age of rapid AI development. We would also be able to make the best of having access to underserved communities to make contributions to improving AI development, such as contributing to evaluations research in a wider variety of contexts, and to diversity in data.

We are currently speaking to AI industry experts to co-design a curriculum that we will then tailor to each charity organisation's needs.

Continuous Support Along The Road to AGI

I would like to use these programmes to provide continuous support to various communities around the world along the road to AGI. Our plan should be informed by AGI timelines, and predictions of how AGI may affect individuals, and collaborate with communities around the world so that we can help each other, and prepare ourselves appropriately for the coming years. (What kind of information would I trust? Fearmongering isn't the best approach, and we would have to adapt to the cultures of different regions and the regulations that are already in place in those regions e.g. Hong Kong would be different from the UK for instance.)

Youth-led Policy Engagement

Through engaging with young women over a period of time to gather their experiences with AI technologies, we could come up with a set of safety guidelines and engage councils(or other relevant orgs) in order to try to include more of young women's needs in their AI-related policymaking discussions.

Scaling Our Impact

If our pilot programme has been evaluated, proven to be cost-effective and a good use of funds, we would like to scale our impact in a way that can help as many vulnerable individuals, young girls and children as possible. The question is, is it possible to proliferate safety as fast as unregulated tools are being proliferated? What kind of information should be proliferated so that children can be protected if an entire series of lecture notes is intractable to mass communicate? What mass communication models have been researched to be cost effective? Is knowledge sufficient or should there be tools appropriate for each region so that each community can quickly proliferate amongst themselves solutions that work for them in the age of AI? If AI proliferators are shoving AI to every scientific discipline, every student, every business operation, how can safety be proliferated in the same way when policy isn't able to keep up? If there was a centralised solution to protect children and uninformed individuals such as vaccines during the pandemic, then gossip has been shown to work well. But what would a simple solution be in the age of AI development? While it does not have to be perfect, does it exist? What are the options?

Should we pay influencers to talk about this and proliferate AI safety knowledge? Who is doing this right now and how effective is this? Should we pay famous celebrities and artists to talk about this and market this? Make an ambassador programme? <todo study research about what celebrity/influencer sponsorship programme has worked in the past.>

Our Approach

We will start by delivering educational webinars to a select number of charities as a pilot. We will engage with these organisations in order to run a pilot programme, and once the impact and effectiveness of the pilot programme are evaluated, we will want to scale our impact by partnering with individuals or organisations with wide reach, so that we can proliferate AI safety knowledge as fast as AI technology itself is being proliferated.

Our programmes are of course open source, and we will create bitesized information that is useful for the day-to-day person.

We loosely reference AImpower.org's Working Model as a starting point for our programmes.

Resources We Need

We need researchers in developmental economics to conduct research on the long-term impact of our pilot programme, as well as recommend ways to monitor impact over time.

We also need AI safety experts to help us design a curriculum of the highest-quality, as well as to update us on changes in the space so that the curriculum can be updated accordingly.