"Future of Life Institute" AI Company Safety Assessment Report 2024

AI Discussions
1

I missed it when it came out but it’s in the print edition of IEEE Spectrum of February, and has been in the online edition for some time.

Leading AI Companies Get Lousy Grades on Safety

A new report from the Future of Life Institute gave mostly Ds and Fs

AI Companies rap sheet
AI Companies rap sheet842×349 37.6 KB

Full report of the Future of Life Institute

The just-released AI Safety Index graded six leading AI companies on their risk assessment efforts and safety procedures… and the top of class was Anthropic, with an overall score of C. The other five companies—Google DeepMind, Meta, OpenAI, xAI, and Zhipu AI—received grades of D+ or lower, with Meta flat out failing.

“The purpose of this is not to shame anybody,” says Max Tegmark, an MIT physics professor and president of the Future of Life Institute, which put out the report. “It’s to provide incentives for companies to improve.” He hopes that company executives will view the index like universities view the U.S. News and World Reports rankings: They may not enjoy being graded, but if the grades are out there and getting attention, they’ll feel driven to do better next year.

I do no foresee the bad outcome of AI is a crapshoot just yet, but trying to check whether the LLM knows too much about biological and chemical warfare is a good idea. Then again, we have officially funded dubious ‘research’ into potentially highly dangerous modded viruses than then go AWOL. Eh, what a world!

Who is on the board:

  • Yoshua Bengio (Professor, CS and OR, University Montréal)
  • Atoosa Kasirzadeh (Assistant Professor, Philo. and AI, Carnegie Mellon University)
  • David Krueger (Assistant Professor, CS and OR, University Montréal)
  • Tegan Maharaj (Assistant Professor, Decision Sciences, HEC Montréal)
  • Sneha Revanur (Founder and President of “Encode Justice”)
  • Jessica Newman (Director the “AI Security Initiative” at Berkeley)
  • Stuart Russell (Professor, CS, University of California at Berkeley)
2 Likes
2

In the iconic handbook called “A Twist of the Wrist”, full of life-saving wisdom, one is about obstacles on the road and fixation of attention. Motorcycle rides to where the driver watches. To survive on the road, the rider needs to battle his instincts and look away from the obstacle into the safe path to pass it. Otherwise even if it is an empty highway with a lone beer can lying on the road, the bike on high speed will likely go straight into that can.

If these brilliant minds concentrate on bright side of things, like a better economic gameplay, we will achieve better score in our limited timespan on this piece of cosmic rock.

3

I wonder if we will see these models use the data that FLI uses to calculate the safety index and iterate to perform better on these calculations as part of their development process and how willing the companies will be to do that.

4

@joshiv - Who / what is FLI?

5

It’s the Future of Life Institute

6

Ah, but that is about a specific situation concerning the eye-hand reflex arc. This situation is not like that. It’s more like sailing the choppy waters and be on the lookout for rogue waves and thunderstorms.

Like this:

robot_future
robot_future611×206 16.6 KB

We also have this, which is from The Verge, but okay:

On the other end of the spectrum, I am probably already unable to get a job because my CV will be trashed by the AI filter looking for the perfect employee even before the HR lady learns about my take on modern sensibilities.