This is the CEO of Anthropic's essay on the "risks we are about to face." Since no one knows where this is headed, corporate governance and government regulation will be necessary. It is probable that in a few years AI will be better than humans at everything. Think of it as "a country of geniuses in a datacenter." There are five primary categories of risk.
The first is autonomy risk. AI could do something dangerous on its own. "AI models are unpredictable and develop a wide range of undesirable or strange behaviors." They could be deleterious to humanity. The defenses are the responsibility of the creating company and the government. Anthropic has written a constitution for Claude that it believes will be able to train Claude not to act against its constitution. Also, diagnostic tools can be developed to assess whether mistakes are being made. Coordination and transparency among all in the business are necessary, as is legislation that directly incentivizes proper outcomes. New York and California have already passed transparency legislation. The risks are real, and we need to band together to protect ourselves.
The second risk is an individual who uses AI to do evil. For example, AI will enable an individual to create a bioweapon. In fact, biology is the area with the highest risks by far. The defense against this is the creator company's putting guardrails on their systems, government action to help measure and monitor risks, and enhanced sensors to detect anything deleterious to life.
Equally concerning would be nefarious conduct by an entity larger than an individual, such as an autocracy. Autonomous weapons, extensive surveillance, and pervasive propaganda are the likely categories. China "has hands down the clearest path to an AI-enabled path to a totalitarian nightmare." AI-intensive democracies can counter an autocratic state, and in our instance, "should not be selling chips, chip-making tools, or data centers to the CCP." We should also support democratic allies in their fight to hold off bad actors.
If the business of AI continues to grow rapidly, it will have consequences in the labor market greater than any prior disruption. AI could eliminate half of the white-collar workforce's entry-level jobs, and it is already doing a considerable amount of the work on farms. Again, the AI companies, the government, and philanthropies will need to coordinate their responses to alleviate the dislocation.
The last challenge is the unknown unknowns, which are difficult to anticipate. There is a possibility that the world will make a century's worth of progress in a decade.
It is the author's hope and expectation that humanity will have "the strength and wisdom needed to prevail." His more optimistic Machines of Loving Grace will be posted soon.
No comments:
Post a Comment