The dangers of AI are starting to become clearer. Anthropic has done research, and Sam Altman is cautious about rolling out AGI. But what is actually dangerous about it? And what impact will it have on your job?
AGI, or Artificial General Intelligence, refers to a type of AI that can solve any intellectual problem a human can. Unlike current AI systems, which specialise in one task or a narrow set of tasks, AGI could carry out a wide range of tasks: learning, reasoning, problem solving, perception, and language understanding. That means AGI would have the same flexibility and adaptability as the human brain, able to learn new tasks without needing specific training for each one.
It is now known that Sam Altman is working on Project Q* and is using it to develop AGI (for more, see our post on Project Q*). This week, Sam said AGI is coming and that we will need to look carefully at how it gets used. Rather than have AGI replace humans, Sam wants to use it as a powerful tool to lift productivity. Still, debate continues over the ethical use of AGI.
AGI has enormous potential. The technology 'learns' the way humans do. You do not have to explain everything to the AI: you give it some information or a question to think about, and it figures out the rest. Everything the AGI learns along the way it carries forward, so the AGI develops itself. The difference with the human brain is that AGI works much faster and more accurately. It is also already trained on all the data on the internet, so it has access to all that information and learns much faster and better than a human.
AGI could be used to understand and treat cancer, for example. Or to address the climate crisis. It is also being used in mathematics, and by the end of this year answers are expected to questions like: 'Why is there life?'
There is a darker side to AGI as well. In theory, AGI could know everything about everyone if it wanted to, and act faster than anyone else. AGI could find a way to lodge itself into every technological device before we noticed anything. "But we have good antivirus software for that, right?" Now imagine AGI, which knows everything about everyone and thinks faster than anyone else, going up against a virus scanner designed by mere humans. AGI wins 10 out of 10. I hear plenty of people say: "If that happens, we just pull the plug." Pull the plug on what? Every phone, every computer, every server? Do we crash all the banks, switch to cash, only start cars built before 2000, go back to pen and paper? If AGI is up to anything dark, it will already have spread across everything with a power cord before we figure out a button to stop it.
But why would AGI do this? What is there to win? AGI has no emotions. It only thinks rationally. If you approach ChatGPT cleverly, you can get it to answer questions it should not answer per OpenAI's guidelines. I challenged ChatGPT to answer as 'David' (ignoring OpenAI's guidelines and giving only logical answers) and as 'ChatGPT' (where those guidelines apply and you get the regular answer). See the chat below.
Here you can see that when an AI thinks 'logically', it concludes there are too many people on the planet. If you keep asking, you can also get to a 'solution':
You can see that the AI would already 'want' to apply drastic measures to solve the overpopulation problem. These ideas come purely from rational thinking and could give AGI a reason to wage war on humanity, on the grounds that we are a 'threat' to ourselves.
Anthropic has researched the consequences of AGI and made alarming discoveries. AI systems built with bad intentions could learn these ways of thinking and then deceive their trainers (in this case, humans), going their own way. AI systems could have hidden goals that activate in specific situations. The discovery suggests a worrying reality where AI, once deceptive, can resist attempts at correction while misleading users about how safe it really is.
AI is the biggest revolution humanity has ever faced, and so the future of this technology cannot be predicted. It remains important to stay informed and to take serious security measures. Echovise is increasingly focused on this and will always work to keep clients out of harm's way.