Op-Ed: Meanwhile, back at AI ‘consciousness’, MIT says it’s a trap. Others might call it an excuse.

The not-very impressive arguments for AI consciousness are generally pretty trite. Whatever benchmark you set for consciousness, there are issues, qualifiers, and objections. What there are not are clear, undisputed markers.

Rumman Chowdhury of MIT has whaled into the consciousness argument with some heavy-duty logic. Apparently mightily unimpressed with claims by OpenAI, Wilian MacAskill, and many others, MIT makes the point that AI “personhood” is also a legal issue.  

This is about to become a gigantic legal issue. In fact, it may already be one, for all anyone knows about any legal issues with AI’s many conspicuous disasters. Determining liability is at the forefront of any possible cases.

It’s also a potential trap of many dimensions, according to Chowdhury. If an AI agent goes rogue, who’s responsible? If the AI agent is a legal person, it is. If not, the entire supply chain that produced the AI agent could be liable.

We’re talking truly unquantifiable levels of liability. This is very much a real thing, and the only question is when it hits the courts. There’s a further trip hazard legally in that if AI agents are treated the same way as employees who commit an act incurring liability. The employer is liable for work-related legal injuries, but not actions outside the job role.

The current situation for AI legal liabilities

The present legal status of AI as non-legal persons is very straightforward. According to Gemini:

Legal liability for artificial intelligence actions rests entirely on humans and organisations, because AI software has no legal personality and cannot be sued. Liability is distributed across the AI value chain depending on where the failure originated.

OK. Let’s get dewy-eyed about value chains. We’ve moved on from talking about legal liabilities to a purely value-based outcome, a sort of reflex of the legal system.

You’ll note that this has nothing to do with AI consciousness, personhood, or anything related. Now add any other legal issue onto this conveyor belt.

See any opportunities for digressing into extreme legal inefficiency and uncharted legal territory that could breed appeals lawyers for generations?

It is a trap, but the trap also springs on the people trying to make excuses by claiming AI is an autonomous person. This is a true bear trap. If you accept that AI is a person, who’s responsible for the actions of that person? The opportunity for excuses can go only so far.

AI personhood can’t be any sort of legally viable excuse in any legal context. The law doesn’t really acknowledge “excuses” for crimes.

Non-human legal persons like corporations are still ultimately controlled by humans. They can’t perform the actions of humans. AI can perform actions, but only at the behest of humans. Humans are the responsible parties, directly or indirectly.

If the AI is faulty or malfunctions for whatever reason, who’s responsible? Under normal consumer law, it’s a product issue. When it’s a legal “person”, it’s a very different and far more complex situation.

The much deeper holes for AI consciousness

Let’s try to avoid the science fiction clichés about artificial consciousness. Asimov did it so much better in his I, Robot series. The proponents are barely recycling his work and adding dubious verbose décor without much, if any, added thought and achieving nothing of note.

The whole idea has more than a few problems from the start:

Does anyone actually need AI with consciousness? Do you want a GPS giving you sales pitches and arguing the merits of designer lattes? Can you think of any possible good reason to have a pseudo-person attached to your basic AI needs?

Humans aren’t too good at managing their own consciousness. What will happen with a totally different class of intellect? Will we have yet another underclass? Or an aloof, remote AIstocracy with no relationship to happy-go-lucky humanity?

Should AI “persons” be treated like humans? It’d be a demotion of sorts. AI is respected more for being AI than being a “person”. Being mass-produced people would also devalue the AI people to the rough equivalent of a box of tissues.

How do you interact with a whole new species? At what points do human interests and AI-people interests diverge? There are so many. This rather tricky dynamic will be the reality. Humans and AI are NOT the same things.

If AI is truly conscious, what rights apply? The “robot rights” thing is anything but new, but if AI has property rights, privacy rights, and everything a human is supposed to have, how does that work? Can an AI take out a restraining order? Can it sue you? When will human rights and AI rights conflict?

How do you punish a conscious AI criminal? Crime is a harmful act, regardless of who or what does it. Consciousness is a characteristic of life as we know it. As a part of the consciousness of an artificial life, or life as we don’t know it, can you punish AI criminals? What’s a disincentive to AI crime? We need to know, since AI seems to be constantly associated with committing crimes.

How do you guarantee human rights in an AI world?

This society isn’t at all prepared for AI law. Human rights can too easily be bypassed unless basic human legal rights are upheld. Whether AI is conscious or not, the whole argument seems to be avoiding that issue.

Laws are based on defining legal injury and applying remedies. Even that’s not happening to any degree. This is a whole new class of law, and it must work.