“The unthinkable becomes possible. Possibilities become reality. Real things become banal and then disappear into the background of our lives.” –Darren McKee
What I read recently Out of Control: The Threat of Artificial Superintelligence and the Race to Save the World Written by Darren McKee. This is an excellent book and my top recommendation for a quick introduction to the risks of artificial intelligence (AI). Here are some of the key insights and messages.
Advanced AI may arrive much sooner than we expect.
McKee begins by discussing the speed of advancement in AI and provides several examples of people who underestimated the speed at which new technologies would spread.
Airplanes are a good example. October 1903, “ new york times He predicted that flying machines may be 1 million to 10 million years away. Two months later, in 1903, the Wright brothers made their first successful airplane flight. Only 66 years later, in 1969, Concorde made its first supersonic flight. It's not just outside commentators who are caught off guard by technological advances. Even Wilbur Wright, who co-developed flight technology in 1903, greatly underestimated the progress, saying, “In 1901 I told his brother Orville that humans would not fly for another 50 years.'' ” he said.
In less than a century, we've gone from a time when people predicted we wouldn't be able to fly for a million years to having airplanes that can fly faster than the speed of sound. I mean, it's incredible progress. It's difficult to accurately compare advances in flight and AI, but by any measure, AI is progressing very rapidly. Many experts and prediction platforms predict that we will achieve human-level AI within the next few decades, if not sooner. The chart below from Our World In Data gives a good idea of the current speed of progress.
Source: Our World in Data, licensed under CC-BY by author Max Roser.
Many domains experienced significant performance improvements in a relatively short period of time. In just a few years, AI has gone from performing worse than humans on many benchmarks to significantly outperforming humans. If you follow the news all the time, you'll notice that these changes show no signs of slowing down. Tools such as his SORA from OpenAI continue to demonstrate new advances in performance.
It might be even more impressive if you actually see it. For example, the image below shows how text-to-image generation has improved over his eight years, going from clearly computer-generated and low-quality to photo-realistic and accurate. It shows that.
Source: Our World in Data, by authors Charlie Giattino and Max Roser, licensed under CC-BY.
This advancement is one of the reasons why many experts and prediction platforms predict the emergence of revolutionary AI within the next few decades. For example, in one of the latest studies, experts assess the likelihood that unassisted machines will outperform humans in all conceivable tasks at 10 percent by 2027 and 50 percent by 2047. It shows that. Notably, the second estimate is brought forward by 13 years from the previous year's estimate. Research shows that progress is accelerating.
AI may change our lives more than we can imagine.
This book argues that one of the biggest impacts of advanced AI is that it will change our lives in ways we can't even imagine. Mackie gives some great examples of how science and technology have dramatically impacted our lives.
Let's think about music. In the 1860s, if you asked people to listen to recorded music, most people wouldn't be able to understand what was being said. It would have been an incredible luxury to have music available at any time, let alone the ability to get recommendations and easily share it with friends. Of course, this also applies to almost everything we have and are doing today. Technologies such as Wi-Fi, transistors, corrective lenses, computers, and mobile phones have always been possible using existing materials, but were not invented or realized until very recently.
Moreover, at some point in time there were always major changes that were unthinkable to people before that time. Our society today is quite exceptional by historical standards. This is mainly due to the Industrial Revolution. If you had lived in an era before this, you would have had very different expectations of life and probably no idea of what was possible. Having machines working for us is clearly an amazing change, and AI could be just as transformative.
There is a very serious risk that AI will make our lives even worse.
This book makes clear that advanced AI is on the horizon and is likely to change the way we live, and makes a convincing case that we should not assume that the changes it brings will be positive. We are doing
Most, if not most, people working on AI are concerned about how AI could harm us. Incredibly, before taking over as CEO of OpenAI, Sam Altman said, “AI will probably lead to the end of the world, but in the meantime there will be great companies.” . As far as I know, this is the first time someone at the forefront of new technology development has said that the same technology will probably destroy the world. The phrase “weirder than fiction” seems appropriate here.
Many experts are clearly concerned. A 2022 study found that almost half of AI experts surveyed believed that the overall impact of advanced AI systems on humanity would be at least 10% likely to be “extremely bad (e.g., the extinction of the human race)” I predicted that there would be %. As shown below, a more recent follow-up study found that expert researchers believe there is a 14% chance that future advances in AI will cause human extinction, or similarly permanent and severe disenfranchisement. I understand.
Source: Katja Grace and thousands of other AI authors talk about the future of AI. arXiv:2401.02843v1, January 5, 2024 (preprint).
McKee acknowledges that AI can and will be used for good purposes, and that AI can be a good thing. But he insists that's not the real point. He convincingly argues that even if we cannot know the future, the balance of current evidence suggests that we should be cautious in evaluating and preparing for advanced AI. We are doing At the very least, we all need to think carefully about what is happening. More realistically, we should also take precautions.
Who should do what differently?
McKee outlines some changes we can make as a society and as individuals.
As a society, we need to take steps to ensure that AI is developed in a safe and responsible manner. To this effect, McKee proposes his eight proposals related to safe AI innovation.
- Establish liability for harm caused by AI.
- Evaluation of powerful AI systems is required.
- Regulates access to computer power and resources.
- Cyber, physical, and human security must be strengthened.
- Invest in AI safety research.
- Create dedicated national and international governance institutions and organizations.
- A license is required to develop advanced AI models.
- Request labeling of AI content.
McKee also suggests that there are a variety of opportunities for individuals.
- Political advocacy: Reach out to government officials about plans to manage AI and demand a realistic and comprehensive plan.
- Media Advocacy: Share useful and accurate AI safety content, support great content creators, and demand more accurate information from sources that spread misinformation.
- Careers in AI Safety: Find a job in AI Safety, leverage guidance from platforms like 80,000 Hours, and bring relevant skills to the field.
- Donate: Support AI safety organizations financially and encourage those with the means to contribute to AI safety efforts.
- Volunteer: Donate your time and skills to AI safety projects, gain valuable experience, and make a difference in the field.
- Join the network: Connect with the AI safety community, participate in discussions, and stay up to date on the latest in AI safety.
- Stay informed: Stay up to date on AI safety developments and insights with podcasts, YouTube videos, newsletters, and forums.
If you're interested in understanding the drivers of advances in computing and AI and their associated societal impacts, follow MIT FutureTech.