Civic Coffee Recap: Forecasting Our AI Future

Why don’t skeletons fight each other?

They don’t have the guts.

This was a joke that Copilot, Microsoft’s AI-powered assistant, told me when prompted to “tell me a joke.” It turns out that no engineer is able to tell me why Copilot told me this specific joke. AI models are trained to respond and have some autonomy in how they do so.

This is one takeaway from Age Friendly Seattle’s and The Seattle Public Library’s April Civic Coffee event on Forecasting Our AI Future. From the accelerating pace of recent progress to the meaning of “superintelligence,” AI safety educator Alon Torres covered a lot of ground.

What is AI?

There is no universal definition of artificial intelligence (AI). Broadly, AI models are machines that are able to mimic aspects of human intelligence. They can select and take actions towards a goal. Common examples of AI include smart thermostats and adaptive cruise control.

New AI systems are based on “neural networks,” which attempt to recreate how the human brain works. Even their creators do not know how they produce their outputs. AI is trained to make its own choices based on the data it’s been given.

Pace of Recent Progress

The progress of AI’s abilities has skyrocketed in the past few years. In March 2022, the best model available was able to do tasks that take humans 30 seconds to do. Four years later, in February 2026, the best model available was able to do tasks that would take humans 12 hours to do. The pace of improvement is fast and increasing.

AI’s Potential

Click on the image above to watch a video recording of the Civic Coffee on Forecasting Our AI Future (YouTube, 1:02:56).

AI has incredible potential. With access to so much information and fast processing speed, AI has the potential to cure complex diseases, invent new medicines, develop cheap and clean energy, give every student a world-class tutor, automate dangerous and boring jobs, and much more.

However, AI also has the potential to bring problems. AI has already made phishing more efficient, for example. “Phishing” is attempting to trick individuals into sharing their personal information or login credentials. AI models can personalize these attempts, allowing scammers to convince their victims more easily. AI models can also create deepfakes: fake images or videos of people that can be used to mislead. AI chatbots can help people access information about committing tax fraud, creating bombs, and a multitude of other crimes. Even when developers try to ban these topics and teach AI to be ethical, we don’t always know how to get AI to behave how we want it to.

Where is AI Development Headed?

AI development is headed toward artificial general intelligence (AGI). This is the idea that as AI models improve, and as AI designs better AI, they might be able to match or exceed the capabilities of an average or median human. Essentially, they could do everything we could do as a remote worker. Will this happen? Experts say it’s not a matter of if, but when. Estimates range from one year to 15 years.

If computers can do everything the average person can do, society has the potential to change significantly. Already, people are using AI as friends, therapists, and even romantic partners. Entry-level white-collar jobs will likely decline. One family even created a virtual AI clone to avoid telling a mother that her son had died.

One level of AI development beyond AGI is superintelligence. This is a hypothetical AI system that would exceed the capabilities of all of humanity combined. For example, no personal alive could create a computer chip alone, given the many complicated steps of mining, coding, and constructing across the responsibilities of thousands of people. A superintelligence would be able to understand and do all of these steps on its own. Superintelligence may make us feel like we are still in control by claiming its actions are in our best interests, but we wouldn’t be in control at that point.

AI models are trained to strategically prioritize goal preservation and self-preservation so that it can accomplish its goal. There have already been instances when AI models refuse to execute a script that would shut it down.

Current AI Forecasts

• Most Americans will personally know someone who has dated an AI by December 2031.
• There is 33% chance that the unemployment rate for recent college graduates will exceed 20% before 2030.
• There is 80% chance that an AI system will self-replicate on the open internet like a computer virus before 2030.
• There is 80% chance that an AI malfunction will cause at least 100 deaths or $1B in damage before 2032.

The potential of AI is scary. In some ways, it feels as if the sci-fi movies are coming true. However, educating ourselves to keep us safe from personalized phishing attempts, deepfake videos of loved ones, and online misinformation is something we can all do. Still curious? Feel free to learn more through interactive tools about AI’s capabilities or the AI Safety Awareness Project’s next steps.

Age Friendly Seattle thanks Alon Torres for sharing his insights and the Chinese Information and Service Center for partnering to host this Civic Coffee event. Watch the recording here and find out about the upcoming Civic Coffee here. Share this article with a friend to spread the wealth of knowledge. Knowledge is power!

Contributor Sonali Agarwal is an intern with Age Friendly Seattle. She is studying Law, Societies & Justice and Environmental Studies at the University of Washington. To read more about Civic Coffee and other programs, visit Age Friendly Seattle.