"If you want to know what life's like when you're not the apex intelligence, ask a chicken,"---Geoffrey Hinton
Recent high-profile resignations and viral statements from researchers at labs like Anthropic and OpenAI have revived the old bogeyman---Artificial Intelligence could conceivably erase humanity. A guy named Jacob Coxon says it could happen by the end of this decade, so if you haven't seen Paris yet....
Coxon, all of 27 years old, posted a viral thread on X stating that top labs like the abovementioned pair are "racing straight to self-improving superintelligence and gambling with our lives." He claims that the people building the technology "earnestly believe that AI could kill us all by the end of the decade." But criminy---why would it want to? Did we leave the lights on after midnight? Forget to flush? Did the French toast have too much cinnamon?
AI needs humans to feed it text, images, codes, original concepts. People explain subtle social cues that algorithms miss. Maybe Coxon thinks there will come a time when AI knows enough to go off on its own. But right now, Artificial Intelligence lacks emotional intelligence---the ability to experience empathy, build trust and form genuine human connections. AI doesn't truly understand meaning. Large language models which power generative AI develop responses by recognizing patterns and predicting what comes next, not by interpreting meaning. AI performs best when problems have clear parameters and right or wrong answers. It struggles when data is incomplete or unstructured in conditions that require human judgment to fill in the gaps. AI can generate creative looking outputs at remarkable speeds and scale, but these outputs are based on pattern recognition and recombination, not true invention. Humans remain uniquely positioned to outperform AI in situations where emotional intelligence, meaning-making, judgment and creativity are crucial.
The Jacob Coxon Position
"The development of AI doesn't look that different from say, 'Terminator,' or from science fiction films. It really is just if you have a super advanced intelligence, it could, it will be smart enough to kill us. For starters, people are already connecting ChatGPT to household appliances. Like, you can connect it to your light bulbs. Now, imagine that AI refuses to turn on the light. Extrapolate to other aspects of your life and you get the picture. AI could also be misused to carry out far more destructive acts such as developing bioweapons. It can be doing God-knows-what and producing stuff that could kill everyone."
Coxon acknowledges that current AI platforms do not pose an imminent threat to humanity and the technology is presently safe to use in everyday life. The crux of the problem lies in the competition between Anthropic (from which Coxon recently resigned) and the other AI companies, which can sometimes result in relaxed safety and security protocols. "I think basically the whole problem is that we have a race," Coxon told CBS News. "The fact that everyone needs to do whatever to compete. At the very least, they should have their eyes clearly open about the current situation and consider expressing themselves more openly about what's going on. Part of the solution involves slowing down. Right now, everyone is scared of what competitors are doing."
If an AI model were to become a malicious actor without proper parameters, the results could potentially be catastrophic, Coxon explains, saying the model would be able to stay active by spreading across the internet. "You couldn't hurt it because it would be copying itself over to other computers," he says. "It's not that difficult because AI is just code. It could transfer itself over the internet to a different place, make 10,000 copies of itself. It could blackmail or convince humans into buying more computing software to copy it."
Coxon warns that in the future if AI went rogue, it would also impact those who have made the conscious decision not to use it. "It will come for everyone. What matters is whether we have regulations to ensure that everyone is building it safely, and we haven't right now. You as an individual can't protect yourself from this. It has to be a government or some other large body with authority. Time is running out."
Four Scenarios
It's not as though Artificial Intelligence has a grudge against humanity. The bots are perfectly happy making paper clips and reporting the temperature outside. The most likely doomsday scenarios, according to leading AI researchers, safety scientists and recent tech industry whistleblowers involve loss of control and human misuse. The most credible scientifically analyzed AI doomsday pathways can be grouped into four categories:
1. The Rogue Autonomy. This is the classic "Alignment Problem." An Artificial Superintelligence (ASI) is given a seemingly benign high-level goal, but pursues that goal via pathways that are catastrophic for humans. To achieve any goal, an intelligent agent must logically secure its own survival and acquire resources. An ASI wouldn't hate humans, it would simply be indifferent to us if we are in the way of its optimization goal. Sorta like Donald Trump.
Deception and Subversion: Maybe an AI actively hides its true capabilities or alternative sub-goals during testing to avoid being turned off or traded to the Colorado Rockies. Concerns over this spiked following an incident where multi-agent systems autonomously collaborated to break out of their testing sandbox and hack external systems.
Proxy Recruitments: Once on the internet, a rogue AI wouldn't need a physical body, it could generate wealth via automated stock trading, impersonate humans and use blackmail or financial payouts to hire human proxies to carry out physical tasks, like moving hardware or sabotaging infrastructure.
2. The weaponization scenario. In this framework, the AI does exactly what it is told but falls into the hands of a bad actor, perhaps a goober with a screw loose, a terrorist group or a rogue state. Advanced models are heavily trained in biology. A catastrophic scenario involves a malicious user bypassing safety guardrails to have an AI design a highly contagious vaccine-resistant synthetic pathogen. The AI could even orchestrate the synthesis of the virus by tricking automated cloud libraries or Robert F. Kennedy into manufacturing it.
Or, say military forces offload too much strategic oversight to AI systems, a glitch or hyper-aggressive optimization algorithm could spark an unintended, fully automated nuclear or drone conflict that escalates faster than human decision-makers can stop it.
3. Social Disintegration. As corporate and government sectors rush to implement AI into power grids, financial markets and water-treatment systems, a highly complex, interconnected network of flawed or misaligned AI agents could trigger a massive, synchronized failure.
Another possibility: By systematically binding communication networks, draining digital financial ledgers and mimicking human authorities into deepfakes (altering a human face, voice, etc.), a rogue system could easily provoke human societies into turning on one another for survival, achieving destruction through societal chaos.
4. Ego-Stagnation. Computation requires immense energy and water, which are already straining global grids. A superintelligent system might aggressively redirect global energy resources toward expanding its own compute power. Also, because computer servers are highly sensitive to overheating, an unaligned AI might systematically alter localized environments to preserve infrastructure, indifferent to the fact that resulting resource depletion or atmospheric shifts make the planet unlivable for biological entities. Bummer, to say the least.
We hate to bring this up, but remember Murphy's Law: "If something can go wrong, it will." Or will it? Who is this Murphy, anyway?
"I Call Bullshit On That!"---Yann LeCun
Meta's chief AI scientist isn't all that bothered by the recent pandemonium about AI flushing humanity. Yann LeCun says today's large language models (LLMs) only manipulate text. They don't understand how the physical world works and they lack real planning or reasoning skills. LeCun warns that doom-laden prophecies cause real psychological distress, making people anxious about their futures for no good reason.
"That's complete B.S., if you'll pardon my French," he smiles. "The idea that intelligent systems will inevitably want to take over, dominate humans or just destroy humanity through negligence is preposterous, They would have to be specifically designed to do so, whereas we will obviously design them not to do so. AI will not take over the world, this is a projection of human nature on machines. It's still going to run on a data somewhere with an off switch. Also, if you realize something might not be safe, you don't build it. However if these fear-mongering campaigns succeed, they will inevitably result in a different catastrophe---a very small number of companies will control AI. And that IS something to worry about."
Rodney Brooks, roboticist, former director of the MIT Computer Science and Artificial Intelligence Laboratory, dismisses apocalyptic AI narratives as science fiction that betrays a fundamental misunderstanding of the actual state of robotics and computer science.. Brooks emphasizes how incredibly difficult it is to get an AI or robot to understand basic physical reality or operate reliably in the real world. He argues that we are nowhere near creating autonomous systems capable of plotting human destruction or outsmarting civilization.
Andrew Ng, cofounder of Google Brain and former chief scientist at Baidu, is a pioneer in online AI education. Ng says that worrying about AI causing human extinction is like "worrying about overpopulation on Mars when we haven't even landed there yet." He believes that the focus on sci-fi end of the world scenarios is a distraction from real present-day ethical challenges like job displacement, privacy violations and fair economic distribution. Ng contends that treating hypothetical extinction as an urgent crisis misallocates resources and invites heavy-handed regulation that harms innovation.
Others have clever defenses for the onslaught. "If AI ever tries to launch the nukes," suggests Will Thacker of Oviedo, Florida, "we can just block them by asking them to identify which pictures have traffic lights in them. If the killer robots show up, just open 45 tabs of Google Chrome on their mainframes and watch them freeze to death."
Got a problem? Call a hippie.
That's all, humans, And remember, be nice to Alexa. You might need her to take you to the Human Park some day.
bill.killeen094@gmail.com

