AI Alignment Theater: Managing AI governance in a levelheaded way
If the recent events of the outages across various institutions in US and around the globe have not already raised the FUD (Fear, Uncertainty, and Doubt) factor on AI governance and alignment theatrics, I am sure it soon will.
While the boardrooms will be busy reviewing the slide decks, approving big budgets for AI governance and alignment initiatives, and debating if the legislation needs to step in; AI will continue to make rapid progress, and similar incidents will increase in frequency.
Although I have touched upon AI governance and alignment in my earlier blogs, I will use this opportunity to reiterate some points and lay out a different perspective on these incidents. There may or may not be a larger conspiracy, state and non-state players timing these incidents, or just plain coincidental events.
Why I see all this as a positive sign?
It is no longer a doubt that AI excels at finding vulnerabilities across millions of lines of code that were too complex or tedious for a team of hackers to identify and attack. AI code generation makes it both easy to identify and attack but also makes it easy to patch. Just like I wrote about ethical AI back in 2018 (https://rnanda.com/2018/09/ethical_ai/). This dance will accelerate securing and patching up of older systems and will improve the quality of software being developed in the future. Sooner, rather than later, AI will run out of vulnerabilities that exist or will not be able to identify any because it would have already addressed them at the time of creating the software.
Why I see this as a bad sign?
There is an old saying in politics – never let a good crises go to waste. The power hungry and control freaks will use these and future such incidents to raise the alignment and the need to legislative controls theatrics. The gullible junta (Hindi word for public) will lap it up because of the FUD factor and happily hand over more power to them without thinking through the consequences.
The biggest issue would be throttling of innovation in AI. The big-tech AI must protect their investors and trillion-dollar evaluations and will eagerly ‘collaborate’ with the politicians to make sure that they have a seat at the table (see my earlier blog on this topic. The attack on open-source and open-weight models will grow and a ban will be asked for. If this indeed happens, it will be an ideal case study on how to shoot yourself in the foot.
How to approach AI governance?
The topic is much larger than what can be covered in a blog. I do plan to talk about it in details in my upcoming online courses on AI. Here are few pointers to start with:
- It is not only an AI problem. It is a combination of traditional software development + AI
- Cognitive architecture is becoming extremely important as part of the enterprise architecture layer. I wrote about it a while ago
- Human-in-the-loop and the role of people cannot be underestimated. The challenge here is that no one truly understands what this means and mistakenly think that anyone can be put into this role, which is certainly not true
When the next incident is splashed by the media and the politicians raise hue-and-cry for desperately needed legislative intervention, I would recommend that we all take a deep breath, a sip of fine wine or whisky and calm down. AI governance needs levelheaded approach, not knee-jerk reaction to every incident.