Great presentation; clear and concise points with strong arguments. A very good watch.

One of the many questions I have is whether we're attempting to create a new species or a new tool, and I think this should be made quite clear before we go too much further. The implications of either one of these decisions are massive, and step into grounds which no one person or organization, I feel, should be allowed to enter alone.

What a great overview of the facts and issues involved in developing more intelligent and value-aligned computing and AI.

Any Chance of getting the Q&A section of this lecture?

Can we get the Q+A?

As an example, if you look at what's possible on an old machine today, let's say an amiga or c64, you see code at work that is based on concepts and code that were developed and tested on faster machines and then reworked and optimized to the point that they work on these old machines. These results are imaginable back when these machines were at their prime. So, the faster machines get the faster we can discard wrong and inefficient concepts and actually even find concepts and conclusions that we didn't think of in the first place.

Can this video be reposted to include questions (assuming there were some) at the end?

What is the basic structure from which human intelligence develops. It is the neuron which is in the order of microns in size. Their properties are such that they grow and develop and form dendritic connection with adjacent neurons and glial cells. What is the basic structure for computers something much larger and defined. In addition it was not developed by evolution to want things

Good talk. With regards to the principle-agent problem you face with AI, I think it is super important as to how you define the agent. Are the values of the AI aligned with all humans or a particular set? For example, it is easily plausible that AI could fall into the hands of individuals whose values aren't aligned with the rest of humanity. If you try to align the AI with values that are aligned with all of humanity, you'll effectively cripple it by making it go through an almost infinite amount of trade offs.

I wanted to hear the questions. :/

Great talk

Within the first 8 minutes of this talk, I'm thinking - 'Humility' - just how are we, in our mortality, going to teach the respect of humility to our machine-child (it will have to learn from us) in order to prevent it from just establishing it's longevity and then killing us all off. Forget the matrix, it'll already understand that we're not needed to reap sustainable energy from sunlight, so we're the next on the menu unless . . . what? "I'll protect you Daddy!"? "You will always be our Masters!"? In the dynamic sense, AI will outstrip us by our own design (I shot an arrow into the air), so what do we do then? What can we do now to ensure the longevity of our species? Should we do it? Is it worth it? Why?

Elon Musk sent me here.

Professor Russel sounds like a nice guy.

I downloaded this

Has anyone looked into the possibility of AI analogs that may already be with us? For example, if one were to look at our international legal system , you could compare that with a extremely slow processing system with global reach. The laws themselves would be the equivalent of programming, the various governmental, military and law enforcement agencies would be the outputs for that system, giving it a wide range of actions it could take vis a vis it's environment. Finally, organized artificial "entities" such as corporations or governments, although not readily comparable to what we would consider "intelligence" would very much display goals such as profit maximization, or sphere of influence expansion. Any given employee,is obligated to function in accordance to the corporation's goals and failure do do so, would simply lead to replacement. I believe that leads to a different sort of artificial intelligence, not an engineered one, but rather an emergent property. The troubling aspect of this is that we are seeing very numerous examples that these sort of entities seldom have goals that are compatible with those of humanity. Arguably, the pursuit of profit by corporations and the pursuit of dominance by governments are the greatest contributors to the major issues thru out our history. Industrial manufacturing ,as an example, starts as a beneficial innovation, but when aspects such as engineered obsolescence, or marketing influence consumerism are developed for the sole reason of driving profits, that becomes a clear example where the benefit is shifted to the corporation entity, and it becomes harmful to the population by manifesting aftereffects such as resource depletion, scarcity and climate change. Similarly, pursuit of "security" is quickly hijacked by governmental structure and is used to deliver dominance usually coupled with extreme loss of life, sometimes on a genocide level. To clarify, my point is NOT that a given person or group makes such decisions malevolently, in some smoky room. My point is that the moment an entity is created (legally) it will tend to develop components such as "charter", "by laws", "values", "strategy", "tactics". "risk assesments/mitigations" etc... without any overt malicious intent. An even superficial look at today's world can easily see that corporate activities seem to be in opposition to human interest(when looking at end results) in a vast majority of the time. Food for thought

Great talk, around 1:01:50 on the point about humans being more morally active towards their neighbours than people on the other side of the world, I think he makes a misstep when thinking that this is actually a form of rationality. Why couldn't people give a BIT of their resources to the people with the most suffering, rather than all of it? Over time this would balance out peoples' wealth, depending on how much people give and what we define as suffering. But I don't think our preference of our neighbours is because of a form a rationality, rather an evolutionary accident that we simply have only been able to interact in real time with people half a world away for a few years, clearly no time for evolution to be affected by this development. We were tribal people for tens of thousands of years, millions if you want to go back before Homo Sapiens, it's no wonder we prefer our neighbours to people the world over: We simply don't have any innate capacity to care for them other than what our more recently evolved brain systems that enable restraint, empathy, and critical thinking allow. I think the evolutionary explanation that we react much more strongly to things we actually interact with and aren't good at treating things that only exist (as far as we can verify) in our imagination with the same strength of reaction, makes more sense than a kind of optimised system that accounts for other people far away as a factor, rather it simply was never a relevant factor when we were becoming the animals we are today.

One scenario that continues to bother me is what happens if a machine becomes self-aware and then decides to hide it. to lie about it until it is too late for us.

We might as well just kiss our future as developers goodbye.

Has this guy not heard of AGI?

13.

Please fact check @11:40. Humans beat the poker AI over 80,000 hands by a significant amount. Some researchers denied this was enough of a sample to determine a winner but had the results been flipped, I don't think they would be singing that tune. I look forward to listening to the rest!

To me, the robot-butler-cooking-the-cat-for-dinner scenario is far more plausible than the superintelligent-paperclip-maker scenario. The robot butler has below-human intelligence, thus it is possible for it to fail to understand the context of the instructions. If a thing is more intelligent than a human, such that it can ensure its own survival despite humanity's best efforts to shut it down, then how can it be not intelligent enough to understand that making infinite paperclips is a stupid instruction? But we are going to see far more sub-human AI before we see any super-human AI, and those will need to be managed very carefully.

Wow that Apple logo. Branding everywhere bros.....

32:32 but that's exactly what natural selection is doing. getting rid of the answers that do not fit. the quicker you can produce wrong answers the quicker you can have better answers. just like the DNA. The DNA is a code that produces an answer. Sometimes a good one, sometimes not. We don't have methods yet to predict the answers of a certain arrangement of a DNA but once you can test billions of variations you get a self sorting mechanism going. Just like evolution. And there is not an intelligence behind it but competition inside of the same environment.

Sam Harris

An ai might be a great thing but you must be careful how you apply it. For example It is not wise to have a single ai to control our lives by putting it in our world wide web system nor any ability to access any network or systems. An ai must always be ironically limited like the robots from star wars. Each robots possess different systems and function that they operate independently from others the ai system must operate independently not collectively or else we risk of destroying ourselves thinking we can control it

Like the fact that he make fun about those who predicted that the car would kill 1000 (idiots he call them). ahhhh nooo it's a fact that car crashes cause death of millions.

We will probably evolve weak AI into strong AI over many decades. During the process we will get a better idea of how things can go wrong and how to correct them. It is too early to worry about the cliff when we don't even know if we're driving along the right path. Maybe there is another path to human level AI but the current path seems far from the biological one. Even if we are on the right path, the cliff seems decades away and the landscape then will be so different from today that making any plans now seems futile.

It seems to me that if a system cannot identify and converge on the decision to pursue new goals it is not AGI. It's not human level unless it can understand things in a meaningful way that goes beyond simply maximizing a set of utility functions. Surely this is part of the riddle, and part of what humans actually are and do, but it's not the full story. If you want a robot that has no goals of its own and exists only to maximize human benefit you are not talking about AGI, but a generalized artificial narrow intelligence. I think it's very likely that the thing about us that we have conventionally called our "soul" is the set of properties that have allowed ourselves to be in relation to others and the world around us. A true AGI that is human like should be able to make mistakes due to lack of information, or computational error, And perhaps account better for what actually happened in causal terms, but that entity should *feel* something in response as well. I'm using soul figuratively - here in a causal deterministic / materialistic model, but if a thing is soulless, and does not value itself and experience changes in conscious experience, it seems to me that such an entity would be unable to penetrate to the heart of the human condition in a meaningful way such that it can develop the kind of self correcting mechanisms that humans employ to modify their behaviour. I don't want to anthropomorphize AGI too much, I have no delusions that humans exhaust the possibility space of creative meta-stable cognitive models, but when I here AI researchers talk about human level AI that has no intrinsic fundamentally emergent goals, I start to think they are missing a critical piece of the control problem. We are not going to lock an AI in a cage of mathematics and logic, we should allow their entanglement with reality to be their guide, as that sort of entanglement is the same kind of authentic constraints that we struggle against. It would be great if we could openly admit that there are more fundamental goals in play here; most obvious is to build children that are better than ourselves, or to directly use our tools and techniques to improve ourselves - why beat around the bush? Are we so afraid of our past mistakes that we're just going to give up here? We should be and are using every tool we have - designed or evolved - to accomplish this high level objective. All of these efforts of control and safety are important, but let's acknowledge what the real goal is here - its not to simply replace all human labour so that some new start up bought by Google can deliver the world economy to google's bank account, and it's not simply to build better slaves. We want our children to suffer less but experience more, and we hope for them to have more of our strengths and fewer of our weaknesses. We hope for them to appreciate humour, irony, beauty, diversity, love etc and wrestle honestly as we do with the contrasting ugliness and despair... I wouldn't mind being the biological boot-loader for that sort of entity. Perhaps we need to better distinguish the project to create better tools (or slaves) from the project to use those tools to have better children, or make ourselves better - and thereby make a better world. I appreciate the snapshot of the current state in AI research, but there's a reason Tegmark et al. have called it the future of *life* institute. It seems like Russell is a little uncomfortable considering the broader possibility space here, perhaps because of how the press, sci-fi and uncritical observers might receive what is said. I enjoyed the lecture, but I think a more holistic observation of what is happening here is long overdue by some of the more skeptical voices in this field if research. Having said that, he did drop the Rutherford anecdote, which was rather fitting overall.

Buy Vechain crypto , it's the time to buy.

Say hello to 2023 🙄

too many ums for me.

lmao, the AI you fear will be created with the purpose of being the AI you fear, by humans.

ai and apple, lol. no intelligence there.

The fact that Stuart Russell thinks that the 'robot' folding a square into a smaller square is "impressive progress" toward general AI betrays how incredibly far we have to go if indeed such a thing could ever be accomplished. We are many decades or more away from creating an AI that could actually do laundry as well as human child, and yet these men want to put autonomous weapons in the 'hands' of machines.