There’s a lot of discussion related to AI nowadays, and while my opinion is just one in the sea of many, I believe expressing your own opinions regarding this is important. AI is becoming a global hot topic, and it looks to be the catalyst for a big change, maybe good, maybe bad. We are seeing this in many fields, from writing to art to music to maths to warfare.
I’ve always been an AI skeptic. I personally do not believe ‘AGI’ or ‘ASI’ is possible. When people are worried about ‘superintelligence’ and how its going to ‘destroy the world’, I tend to disagree with them. Sentiment on ‘superintelligence’ is rising, and more people are believing it will bring the end to humanity. While I would love to discuss why I believe its impossible, I feel the discussion is moot. Instead, we are already seeing AI being used to do bad things, even without ‘superintelligence’.
There is a great video on youtube by the content creator LEMMiNO titled Consumed by the Apocalypse. In it he goes through 8 different scenarios on how the apocalypse could occur, 4 natural causes and 4 human driven causes. The 4 human driven ones are:
- Unnatural Pandemics (a human made virus)
- Nuclear Holocaust (If our current nuclear weapons were to be used)
- Artificial Intelligence
- Environmental Decline
When I first watched the video, I believed (and still believe) the most likely is Environmental Decline, with Artificial Intelligence being the least likely cause. My reason being climate change is such a complex problem, and requires a lot of cooperation between different governments, parties, nations, and people. The effects are long lasting, take a long time to notice, and slowly but surely whittle away at humanity. Things that are related to climate change are also essential to people livelihoods; energy generation, food, transportation, land use, the economy, etc. I have a hard time seeing how we solve all of these issues and prevent environmental decline that could wipe us out.
Artificial Intelligence is not like the environmental catastrophe. AI is not essential. It is an additional tool that has its benefits, and we can ‘easily’ pull the plug whenever we want.
The question on whether or not AI is ‘just predicting the next word’ or is ‘thinking’ or is ‘sentient’ is a philosophical one. I lean towards ‘just predicting the next word’, as that is really what an LLM is. AI is fundamentally different from a human, it does not have emotions, a physical form, a conscience, a family, or the hundreds of other things us humans have. I acknowledge there can be debate regarding every single trait a human has and how those traits make us similar to an AI; but despite that, I will state strongly that humans are different from AI, and no matter how much we argue about how similar AI and humans are, I believe we should value ourselves much more than we value AI.
This leads me to my main point of this blog post. The output of AI is not powerful. It’s not important. It won’t ‘turn evil’ and eradicate us. The only way AI can do that is if we humans allowed it to, by treating its output that way. It’s not valuable by itself.
Here’s a hypothetical: Could an AI launch a nuclear bomb to eradicate humans? Some pedantic people might say yes with some convoluted thought process, but clearly the answer is no. The only way it could do that is if the people who currently hold the power to nuclear bombs gave the AI the power to launch a nuclear bomb. But a two year old could also launch a nuclear bomb if we gave them the ability to. The critical point is not the AI, its not the two year old, its the people who currently have control over this stuff.
This means as a society, we need to push back on the use of AI on our critical systems. We cannot have the AI control our nuclear weapons, combat drones, missles, etc. We cannot have it be the sole manager, doctor, or judge. All of the most critical decisions should always be made by a human, and the one bearing responsibility should always be a human. If AI nearly causes a war between the superpowers of the world, then the one at fault is not AI, its the humans in charge. They should be convicted as a human, by a human, and sent to human jail.
– Begin partially ranty portion
I don’t even think this is a controversial opinion, but we’ve seen it be ignored multiple times already. The OpenAI hugging face incident is one that makes me so angry. Someone or even a group of people should be punished for it. OpenAI should be hit with a big financial punishment. But instead the exact opposite is happening, where no party is being charged not for even negligance. Instead, OpenAI are treating this as marketing, get no punishment, and instead they have it happen again. Not only that, other companies are now openly saying their models did it too.
– End ranty portion
We need to develop an international agreement which ensures AI is not given control of these systems. AI can never be responsible, so if something atrocious happens, the humans in charge must be responsible. Let’s start ‘small’ with the usage in war. Let’s make it a warcrime to place AI as the sole user of weapons. The only ones that should be allowed to fire weapons at the enemy should be a human. Always have the trigger be a human.
While this proposal may sound as optimistic as coordinating countries to prevent the climate disaster, it is not. This is simpler. Like I said before, AI is not essential. If we know something is bad and not essential, getting people in power to prevent it is more doable (although it can still be hard, such as with nuclear weapons). But the point remains the same, the problem with AI is not the AI, its the people in power trusting and valuing it too much, and us not demanding responsibilty when its misused.
Footnote
This blog post was meant to be much longer, with me expressing my opinions on AI in education, art, writing, social media, research, academia, security, etc. However I ended up getting some writers block when I finished the education portion. I want to spend some time reading more and collecting my thoughts, but it will take time, so I decided to release this one on its own. I probably will release the other parts separately as well, as there is a lot to write about.