The Alignment Problem
I have a confession that won't be surprising to anyone who has been around me in person.
I've been getting distracted by AI lately.
I can't tell if thinking about AI is Resistance or critically important.
Right now, I'm thinking it's a little bit of both. I think that it's critically important to think about these things, but I also think that I tend to worry about it at the wrong time.
One of the things that concerns me is how most people don't seem to even be aware that something big is happening.
Big isn't the right word.
Something revolutionary is happening.
Even as a type this, revolutionary doesn't seem to do what is happening justice. Revolutions have happened before.
I'm trying to wrap my head around it all, but it's still too big. Every rabbit hole of if it can do that, then it means this will happen leads to another revelation that blows my mind.
Why am I playing with it? What am I trying to figure out?
I'm trying to figure out what it is capable of. The answer is that it is capable of more than I would have imagined.
What am I playing with?
I'm playing with the AI tools made available by an organization called OpenAI. Their most well known tool is ChatGPT.
Here's the exercise that I've been having people do when they come to the farm.
I have everyone sit down and start prompting it.
Most people start in the same couple places. The most common is some sort of joke.
Then I push people to ask it about a problem that they are actually dealing with.
It is possible to make it err.
As many of you have found out, it is also possible to be blown away by what it is capable of. We have received excellent responses to prompts relating to:
- Forms needed for disolving a 501c7 organization in PA
- Nymphing patterns for fly fishing the Loyalhanna
- Nutritional deficiencies associated with a specific condition
- Drone models capable of creating orthomosaic maps
- A story about a woman with a farting problem
- And many more
Yes, I have made ChatGTP tell me fart jokes. For the first time in my life, I have a fully captive audience that has to respond to my madness. My entire family has a new "character" in our lives named Grandma Farty.
I'm not sure, but I wouldn't be surprised if I would make a leaderboard for "Most prompts related to farting" and "Most prompts ending with answers of 'I'm sorry, I can't answer that because people might find the answer inappropriate.''
Joking aside, I still don't think people realize the significance of what is happening.
If you go to the OpenAI website, one of their main links is for a project they call "Alignment". In the AI world, alignment is the term used to refer to the challenge of trying to make sure AI is built in a way that serves humans.
There is a video on the OpenAI Alignment page.
I would encourage you to go watch that video. I'll link to it at the end of this discussion.
For now, here is a transcript from some of the most important parts of the video.
Remember, this is from the organization that is responsible for some of the most advanced, publically-accessible AI on the planet.
Very quickly we can end up in a place where machines are far more capable than us...so whats really interesting is as the AI systems get more capable, they don't automatically get better at doing what humans want. In fact sometimes they become less inclined to following human intentions. This is what we call the alignment problem. I think solving this problem is of critical importance if we want life on earth to go well.
Here are some additional notes from the video:
The system learns incorrect things.
The system has to be incentivized to tell us the truth, otherwise it can be dishonest.
AI can create content that is indistinguishable from what humans can produce.
AI is good at poetry, writing, music composition, and science problems that are beyond the capabilities of a human to solve.
This is not a trivial problem.
As AI becomes more powerful, alignment becomes even more critical.
It is probable that AI systems will become part of everyday life.
Here's the part where I think things have the potential to get problematic.
In order to solve the alignment problem, it seems that you have to give the machine a value system. You have to teach it right from wrong. It's kind of like raising a kid.
Who get's to make these decision though?
As these thought machines continue to penetrate daily life, whose deciding on the value system?
Is there room to appreciate how my context may be different from your context?
Or does the model rely on consensus that will end up hurting some people and benefiting others?
What happens to dissent?
Do you realize how easy it now is to flood the airways with a single "approved" viewpoint while erasing all dissenting opinions?
These discussions are going on all around us, but they are so politicized that most people are turned off by all the bickering, myself included.
Let's try to wrap this up for now.
Competant AI is no longer a future thing.
Competant AI is right now.
AI seems to grant an unprecedented amount of power to the people who control it.
I don't think this is something that we can afford to throw up our hands on and say that it's beyond us.
There are three suggestions that I feel confident in making at this point:
- Be especially wary about any apps that use AI to capture your attention and manipulate your brains dopamine response to create habitual app use. Get off the app and take your attention back while you can.
- Spend more time in nature. Go on more walks. Take time to cherish the here and now.
- If you find yourself inspired to pass judgement on others because of some view that seems at odds with the currently in vogue meta-narrative, take a step back and choose to love them anyway. In general, choose to love more.
What do you think?
I know what I think.
Life is good!
Here is the link to the page with the alignment video:
OpenAI
Here is the Grandma Farty prompt and story:
Not bad ChatGPT, but I can still do better!