Jordan Hughes
For a different perspective, see my blog piece in Artificial Times: "There's no cause for alarm - just be quiet and drink your Kool-Aid," written in response to an article by IBM researcher David W. Buchanan published on Feb. 6 in The Washington Post: http://artificialtimes.com/.
Dear Jordan,
I have checked your profile, and saw that we have common interest areas, although different perspectives: I think we (human common sense) have no rational view even for the next 10 years, not for 10000... maybe you would like this lecture of mine.
Regarding AI, congratulations to your very professional blog, it was a great pleasure to see – I hope one day I will appear on it :-) I have checked some of your links, and I think our perspective is not that different.
It seems that we agree on the existence of danger emerging from IT systems. The difference is that I focus on the human side: we should not rely on any ethics from an IT system, we only get back what we put in. This is independent from the speed and performance of the system, and already causes huge problems. With more computer power, it will surely increase, and this is lethal to our civilization unless we change our fundamental motivations and attitude.
We also agree on the fact that the lack of fundamental definitions (intelligence or consciousness) makes objective evaluation and discussions fuzzy. However, I have found practical definitions that I actively use in my system design and programming tasks. I plan to write about them soon.
I still disagree in the “automatic emerging” of a conscious entity by any definition, from the mere performance increase. I think for any improvement it is fundamental to find a different IT representation of our knowledge than the current business-driven approach, separated by the various platforms, languages, formats and toolkits. Without a completely transparent and reusable knowledge store, the more an IT system “knows”, the greater amount of custom source code and configuration is created, therefore the system is less reliable, flexible and interoperable.
I think this is the key step to forget the current hordes of closed ANI systems: when their data structures and messages become transparent, they can form an interacting mesh network. I think this is the base of an ASI, and its “personalized agent”: AGI. No wonder: the focus of my research is a transparent knowledge representation, I have quite advanced working solutions – and not less ambitious plans :-) [acronyms: Artificial Narrow/Generic/Super Intelligence, as explained here]
The other difference is how we see the coming massive power of our global IT infrastructure, and the articles I read show the same fear that I mentioned in the “physical power” analogy. Before the “machine revolution”, people were afraid of their imagined “inhuman” power – today we are not afraid of a crane, an airplane or a power station, although they are much stronger than those people could imagine (the difference between the person from 1750 and 2015). I think the fundamental difference is a paradigm shift: we don't have to compete with machines where they are better, but can form a kind of symbiosis, combining our true human values with their not less true artificial power.
This symbiosis already works for me: a part of my mind is not in my skull, but exists on the Internet and the word processor, design and development tools I use. Without them, I could not do my job at all. A system that I build “has” all my knowledge that I collected in months, but some concepts literally took me decades to achieve. My system contains and uses it all in every single moment, does more in a second than I could do in weeks (but I would make tons of mistakes). Is this system smarter than I am? Of course. Is it dangerous? Maybe, in wrong hands – but not because it is smarter or faster. It is me who adds the “human part”: the roots in ancient philosophy and religions, and the motivation to seek for the benefit to our civilization.
Yet another fundamental difference: immortality. I guess someone envisioning human immortality does not understand the most fundamental aspects of life and human thinking. The very definition of all life forms contains death as the rule of the game – and the most fundamental motivator of human thinking is knowing about and hopelessly struggling against this concept. There is a clean analogy of “biological immortality”: cancer, and this analogy perfectly describe our attitude and its result on our living environment. For human beings, this is the key improvement required to enter the new era: understanding and embracing personal biological mortality – and “memetic”, informatic immortality: our ideas, or practically anything appearing about us in the IT infrastructure will be preserved by it and become “timeless” compared to our limited and time-bound existence. Another example: space is not for human individuals as the current “spacey-tales” suggest, but for the whole human race, through lots of personal sacrifice, pain and death. This is the only way our civilization improved in the past and will improve in the future.
We'd better grow up to this before it gets too late.
Alexandru Grosu
It is time to implement the four laws of Asimov into all the AI APIs. Now, it is still safe.
The most fundamental statement regarding the AI is that there is no way to implement ethic concepts in algorithms and no reason to assume that a software can be ethical. By the way, Asimov himself gives a very foggy explanation about how the laws are burnt into the pozitron brain.
I think the explanation is simple: his robots are not "AI previews", but represent human beings with greater power or higher level mission. His stories show psychological effects of coping with greater tasks or fundamental ethical laws, racism, fear. He talks about how our intellectual, political or financial elite should behave, what laws they should follow after they have separated themselves from the simple burdens of human life (the lower section of the Maslow pyramid if you like).
I also think that his laws don't work, because they are negative or conflicting: “may not... except”, just like “don't kill” or “don't be evil”. They freeze a conflict between our motivations, rules and interests: it would be profitable to be evil, but I should not be – but hey, what is evil anyway? and how far can I go without being evil? should I lose if my competitor is evil?
I think positive laws are harder to follow but they focus on the path where you should go instead of walls that you should not break. I see this improvement between the 10 commandments “don't”s and the simple “love your fellow as yourself” (if you love someone on the same level as yourself, there is no motivation to break the commandments...) My actual favorite is the Shotokan Dojo Kun (which is absolutely useful and practical in everyday life as well):
- Seek perfection of character
- Be Faithful
- Endeavor
- Respect others
- Refrain from violent behavior
And again: this is not for the AI, but us who create and use it. We failed on this mission with the machines (“officially sacred” weapons cause the same suffer), and true: with AI, this failure can easily be our last one on civilization level, but not because of the “evil AI” but our own irresponsible, childish attitude that now meets with Darwin's law.
We are not "too big to fail", but "too big to survive this failure"...
