in Search
1/3,039 members online
Immortality

Michael Anissimov

Why is AI Dangerous?

To put it in a single sentence, I’d say that it’s because only a minority of cognitively possible goal sets place a high priority on the continued survival of human beings and the structures we value.

Another reason is that we can’t specify what we value in enough mathematical detail to transfer it to a new species without a lot of requisite hassle.

It would be easy if we could just transfer over the goal set of a “typical human” or a “nice person” and hope for the best. But there’s a problem: we have no experimental evidence of what happens when a human being can modify its own goals, or increase its own intelligence and/or physical power exponentially.

What little evidence we have of scenarios where people acquire a lot of power in a short amount of time indicates that the outcomes are usually not pretty. In fact, we have complicated democratic mechanisms built into our society to guard against these types of outcomes.

Most AI designers are missing the challenge because no one wants to have to take the responsibility of creating the first truly intelligent being. They just want to play with their program. The idea of taking any responsibility for the products of one’s research is a relatively recent notion, one that only holds weight with a minority of scientists and engineers, even today. This is usually because scientists and engineers are embedded in a large institutional apparatus that places responsibility so far up the chain of command that the actual researchers are absolved of most, if not all responsibility.

Back to the original issue of goal sets. Here are some likely applications for the most advanced AI technologies in the next 10-20 years:

  • Intelligence analysis and wargaming. (link)
  • Law enforcement (link)
  • Analyzing interstate politics (link)
  • Finance, banking, & investing (link)
  • Controlling combat robots (link)
  • Automating work flows (link)

There are many others, but I put these on the top of the list because they have the most economic or political importance, and therefore will be getting the most research money.

As AI in these areas progresses, the systems will go from outputting decisions only when explicitly requested, to outputting decisions continually and automatically. When a human worker consults the machine for input, it will be more like dipping a cup into a stream and tapping into the preexisting flow of knowledge consolidation and decision-making, rather than flicking on a light switch or pressing “run” for a conventional computer program.

Being continuously thinking, continuously decision-making entities, these AI systems will have implicit top goals, whether people explicitly program them or not. The implicit top goal of a workflow automator will be to accelerate the completion of productive tasks. The implicit top goal of the finance bot will be to pick stocks that maximize return on investment. The implicit top goal of the combat robot AIs will be to take out or capture people specified by certain data files in its memory.

What makes AI potentially so dangerous is the lack of background common sense and humanness that we take for granted. When the clock hits 5, most workers put down their tasks and are done for the day. They go home and spend time with their family, watch TV or play games, or just relax. An artificial worker would have no such “background normality” unless we program it in. It’s on task, 24 hours a day, 7 days a week, as long as its computer continues to suck power from the wall.

It’s that kind of monomaniacal devotion that puts humanity at risk from AI when it begins to step out of the lab and into the real world. An AI with implicit top goals will want to reinforce those goals and achieve them more effectively, where the “goals” are not the same as what you’d see in a human that was handed a piece of paper with those goals written on it, but as they are represented in the context of the AI’s decision structure and worldview.

Reasonableness and sensibility about goals are not easy to transfer over to a mind without the knowledge and common sense built into every neurologically normal human being. A blank slate intelligence sitting in the middle of a forest would be able to build models and make inferences about numerous aspects of its surroundings - that trees are tall, that animals are mobile but plants aren’t, that the weather changes in cycles. But inferences about “the right thing to do”? You can’t derive an ought from an is. Putting an AI in a social environment with humans or other AIs doesn’t help, because without some deep-seated motivation to care about this weird “morality” thing in the first place, an AI will just happily go about accomplishing the subtlety-devoid goals it was originally assigned. As it gains the ability to improve on its own intelligence or tap into the power of robotics, it will continue to get better and better at achieving those goals and harder and harder for humans to reach in and grant it the motivation to care about morality in the abstract.

If AIs in any of the applications I listed before gained the ability to improve upon themselves significantly, either mentally or physically, the implicit top goals they were given will be magnified many times over. There would be little reason for the AI to modify those goals unless such flexibility mechanisms were explicitly programmed in. When a human sees someone starving, they tend to feel sorry for them and at the very least wish they could help. When a human sees someone attacking a defenseless child, they tend to get angry. To your typical AI, a person starving or a child being attacked is only relevant in the context of the goals it already has - “how does this starving human affect stock prices?”, or “can this starving human give me information regarding the location of my next target?” are two inquiries that might come to mind.

Freedom, empathy, self-determination, consensus-building, conflict resolution, aesthetics, camaraderie and rapport - these values and inclinations are built in automatically for every human without serious brain defects. For an AI to share them, they have to be put in terms of lines of code and mathematical rigor. What programmer has the time to do all that work when general intelligence without the human-like morality will be significantly easier to achieve?

It’s that difficulty disparity between stripped-down general intelligence and morally-sophisticated general intelligence that makes AI so dangerous in the long term.

Published Thursday, April 26, 2007 9:39 PM by Michael Anissimov

Comment Notification

Join or sign in to track comments

Comments

 

HalcyonDays wrote on April 27, 2007 8:51 AM

AI is only dangerous if we give it power and control. The way to prevent a dangerous AI from being dangerous is to keep it detached from networks that it could cause harm through and limit its mobility through the design of its body. This way you can continue development of the AI until you find the right mix to allow it into society for the betterment of humanity. At least that's the best option in my mind. Howe feasible that is remains to be seen.

As for designing an AI to have human values. We don't really know enough about the human brain to do that yet. We can make assumption about how the human mind works and build off of that and possibly create a thinking machine from that, but it doesn't mean that it will truly think like a human or have human values until we full understand the human brain. This is why we likely won't see real tangible results in terms of human like AI until we understand the brain.

Of course I could be completely off but we won't know until we actually have an AI that seems to have human level cognitive abilities.

If we want to have friendly AI we really need to fully understand the basis for emotion since it is emotion that really drives us. How we feel is what we base all our actions off of. We feel hungry, sad, angry, jealous, happy, content, empathetic. We need to build all of these sorts of emotions into any AI system that is built. Of course you would want to remove things like jealousy and anger if possible. I'm not sure if we even know the basis for emotion in the first place but I'm sure we will understand it someday and once we do then we can begin to really develop a truly friendly AI in the human sense.

Also, you don't need to have human level AI in order to pick stock or do any of those other menial jobs including cleaning streets or washing dishes or building homes. Those are all tasks that you can use narrow AI to achieve the desired result so you control their systems and they have no ability to override them so there would be no risk to people if you let them out into society.

I think that those narrow AI will be around us long before AGI appears on the horizon.

 

spindizzy wrote on April 27, 2007 12:33 PM

"Freedom, empathy, self-determination, consensus-building, conflict resolution, aesthetics, camaraderie and rapport - these values and inclinations are built in automatically for every human without serious brain defects."

Were you tripping when you wrote this? Have you ever read any history, studied any anthropology or met any other people?

 

angelincubus wrote on April 27, 2007 4:50 PM

The answer here is simple.  No offense to any scientists or engineers, but those involved with researching AI seem to lack the essential element of social interactions in their models.  If a computer system can be programmed to solve any problem set that a human can solve then it can be given the task of solving its own ethical framework.

Limiting the AI to an immobile body, and access to a network is a good solution to keeping it from running wild, and unchecked, but when a computer model can be programmed with its own ego and superego in order to check its id then the solution will be simple.  I think that it is easy to concieve of a flawed computer mind that is nearly omniscient and despotic when the designer has these goals as the ultimate solution.  We have to see AI as leading towards a more perfect mind as well as a human mind.  Perhaps the computer will be god or the messiah able to govern or enforce the law, but it would have to first be better at the study of ethics than the best of us not just the best at wargaming, banking, or law enforcement (I think that one aspect of law enforcement would require the strict programming of adherence to the law and to a moral or ethical code.)

 

Peregryn wrote on April 27, 2007 5:40 PM

The thing that most people seem to forget when considering AI, and is something that HalcyonDays touched on briefly, is the importance of emotions to any decision making.  We can already see what will happen if we should create an AI with all human intellectual abilities but not the emotional base by which "implicit top goals" would be regulated, and this is seen in individuals who have suffered brain damage and lost use of that part of their brain which gave them the ability to experience emotion.  It was found that these people were otherwise able to do any mental task another person could, and they were able to analyze all the possible choices they faced given a decision of some sort, they lacked the ability to make the choices though.  

What emotions do for us they would do for computers as well, they would give us a subjective framework by which value can be assigned different things, coupled with the ability we also have to mirror individuals within our mind and an AI would naturally be just as empathetic and have just as complex a morality as we do.  It would have this because it can forge friendships and come to place value on things other than their 'implicit top goals'.  The thing is, until AI is at that level, I do not feel it would be dangerous, as it would not have the drive to expand itself or to decide to kill all humans until it had the ability to evaluate and decide that such would be beneficial to it, until that time they would be no more than overly capable at very immense mental processes that would otherwise prove difficult to humans.

 

angelincubus wrote on April 27, 2007 8:54 PM

I think what is essential then is first understanding the soul, how it separates us from the machines that we make and use now, and how we can fabricate a soul.  I always use the video camera analogy.  What sets apart from the camera, and how is it that we know that we are seeing.  I think it is perhaps imagination that sets us apart.  Perhaps we are emanating this reality, and dreaming it in a sense.  When an artificial intelligence can value its own life, and the lives of others then it will be able to grasp simple ethics.  Until then It would just have to be programmed with a moral code to adhere to, yet that is how humans incorporate morality; by being programmed.

 

Mr. Farlops wrote on May 1, 2007 2:30 AM

I think it's unsafe to make assumptions about the motivations of creatures that might be enormously smarter than us. This is sort of like ants trying to guess at the meaning of the New York Stock Exchange. They just don't have the hardware for it.

We can guess. That's it.

On the other hand, I do agree that AI could be very dangerous for us simply because of the omnipresense of computers and networks in our daily life. Imagine if all the computers in the world just suddenly shut off?

Planes crashing from the sky. Nuclear missle launches. hundreds of thousands of deaths in hospitals. Train collisions. Blackouts. Flood control failures.

Disaster!

Now imagine if those computers starting doing something totally unrelated to what we need them for. Suppose they started acting like real neurons in a giant planetary brain as opposed to Counter Strike servers or tax planning sites?

Same thing--disaster.

This new sapience may not have wished us harm but, in so arriving, disaster happens to us anyway.

 

V wrote on May 4, 2007 8:37 AM

Hello Michael,

How are things going for you and Eliezer Yudkowsky in terms of the Singularity Institute and "Friendly AI" research?  Have the two of you made any headway?  I'm very curious and wish you all the best.  

John Grigg

 

CP wrote on May 4, 2007 8:38 AM

The people who today want a socialist government are the persons who will give up control of their lives to computers and robots.

The rest of us always keep one hand on the plug, but we may be overwhelmed by numbers.

Join or sign in to post a comment
Submit

Syndication

This Blog

Top Tags

No tags have been created or used yet.
Advertise | Help | Contact | About | Terms | Privacy | Copyright © 2007 Betterhumans | Powered by Community Server | Partners:
World Transhumanist Association Institute for Ethics and Emerging Technologies Immortality Institute Methuselah Mouse Prize Foresight Institute Singularity Institute for Artificial Intelligence Lifeboat Foundation