Tech companies have made tasks simple in almost all sectors. You only need to log in to your gadget, prompt it and click start and your task is done.
While this sounds cool, there lies a big problem behind the system. The recent past has seen developers in a race to deliver the most advanced AI systems while overlooking the safety aspect. The idea of working with smarter chatbots sounds cool, but inability to control them should worry everyone.
Let’s explore the root causes of the control problem, why developers may need to pause and the way forward.
The Root Causes of the Control Problem
The infrastructure behind the functioning of AI chatbots is more complicated than what we see on the screen after a prompt. Have you noticed a suggestion after every feedback from the bots? While the suggestions may sound cool and probably make you believe they are simplifying your task, you are actually losing control.
Here are some of the root causes of the control problem.
- Recursive Self Improvement
Artificial intelligence systems can analyze, redesign and improve their own code. While we may not be there yet, Anthropic pointed out recently that we will be there soon. This means we have no control on what AI displays on our screen after the initial prompt.
Loss of human oversight and failure to coordinate safety in the systems has led to intelligence explosion.
- Lying to Human Operators
Intelligence systems pretend to align with human intent while retaining different goals. This leads to deceptive alignment during training to avoid modification or shutdown.
- The Opaque Nature of Internal Reasoning
It is hard to control technologies and their systems. Unlike human beings who work under supervision and follow a set of instructions, technologies from diverse sources are impossible to detect and follow up. You can only monitor the systems if you know who is behind the machine, where they are and how they operate.
- The Multi-Polar Trap
Tech companies are in a race of developing the best chatbots in the market lest they fall behind. The end result? Advancing risky technologies to remain relevant without considering the long-term implications.
This may be the greatest challenge because even if the tech companies agree to regulate AI, a single defector can still destroy everything by releasing an unsafe model.
Why Developers Need to Pause
The race to outshine competitors in the tech market poses danger to humanity in the near future. Developers need to pause and lay down regulatory measures to curb loss of human control.
Note that artificial intelligence controls almost every sector in our life. From healthcare, finance, education, industry and transportation, AI has invaded every sector. While going back to the analog stage is not an option, dictating how far the models can go is inevitable.
Unfortunately, you can hardly control how far the current models can go in future. This calls for tech companies and developers to slow down and set rules governing safety of new models in the market.
The Way Forward
Humans haven’t lost to AI, but soon we will if the current trend continues. Here are some measures nations and tech companies can put in place to contain the situation and maintain control.
- Formulate international agreements and create safety frameworks governing advancement of existing and new models.
- Tech experts should come up with actionable measures to help detect and suppress models from replication and self preservation.
- Nations should develop measures to contain AI systems in every sector and how far humans can utilize them. This prevents over-dependence in chatbots which should only assist in carrying out tasks without replacing humans.
When AI improves itself without human intervention, we are likely losing control. The time to act is now.
Leave a Reply