Creating Human-level AI: How and When?

[@thearbiter302]

I would enjoy seeing debate style panels between the panelist.

[@carloscervantes836]

thearbiter302 I would like a round table casual and natural back and forth conversation with one of them just reading out question submitted from multiple sources, including college students in the relevant fields. Read a question, let them all discuss it until no comment is left and then move on.

[@AdeelKhan1]

In addition to a simulated world that the agents can train in. My uneducated take is that the following may also be required. Including, but not limited to: a) Traditional CPU and GPU powered architectures may not be enough. So we will need newer forms of computation. b) If we design a fundamentally different type of architecture (beyond Von Neumann's designs), then we need to figure out how to make memories accessible in a relatively seamless manner. Here, you do not want all the memories to be queried on a continuous basis. That will just overwhelm the agent. So, the memories have to be encoded with different layers of sorts. Contingent upon the task being performed, the agent will access different set of memories. There would then be modules that the agent will make use of, depending upon the kind of task that the agent is performing. | I think, the current approach looks something like this : Data is in a database and neural networks are trained (supervised or not) in order to solve a particular set of problems. If we rethink the ontological structure of how the data is stored in the first place and then work towards creating modular components. Then these modular components would then select the kind of pattern recognizers that they need to invoke and the pattern recognizers will access the memories that they need to access. I don't have a background in compsci, so I do not know what I am talking about here.

[@Booklamp53]

what did the first three people think about the first question?

[@top1percent424]

booklamp53 Exactly. They should've been asked first.

[@adrienrenaud3630]

Quoting Bengio on the when: "I don't think it's reasonable to make these predictions" kzbin.info/www/bejne/kHm8iaNpnK92l6s

[@DistortedV12]

Jurgen Schmidhuber, Yoshua, Yan Lecun, Demis Hassabis are the ones here; they just need Geoff Hinton

[@cityinmind]

Why do they have them sitting so close to each other? a few of them seem uncomfortable and trying to make sure they dont touch the person next to them

[@aaronodom8946]

cityinmind They're all social-awkward, nerds. lol Just like the ones we all remember from high school.THey rigged the chairs close together on purpose; so maybe they'll slip up and say something too secret!

[@swapanjain892]

This is not the full upload.

[@michaeltinker240]

Super-human AI is already here, and its name is Demis Hassabis. What a legend.

[@mo.kashefi8744]

Was there a first part to this video that was chopped off?

[@delr853]

Personally, I think the first AGI or ASI will be developed by an AI through a model that mimics evolution (AI creates a better AI, then that AI creates a better one and so on). Google is already working on something similar to this concept. As for when, I think Kurzweil is correct, we keep thinking in a linear fashion. We need to learn to think logarithmically. Any estimate longer than 20 years from now is just simply wrong.

[@gcgrabodan]

One the other hand: We kept overestimating progress in the field of AI for 50 years. We should also learn to learn from our mistakes, meaning AI might come in 500 years ;-) It is called the planning fallacy. So, how do we combine this inside with the logarithmic thinking? I dont know...

[@martagutierrez3858]

Del R, could you provide a reference to the Google work you speak of?

[@KaplaBen]

Tenenbaum should have been among the panelists

[@AdeelKhan1]

41:31 Agree with the two points that Guru Banavar has raised and that have concluded around the 41:31 minute mark. Working my way backwards, I actually asked a general physician here in the GTA (Greater Toronto Region) if they have machine learning models assisting the doctors and nurses with reading data from FMRI and ultrasound scans. The answer was not yet. Regarding the other point, it seems like a ton of evolution has occurred in the domain of modelling neural networks. So, even I, from a non-software dev background can appreciate the kind of changes that will be required.

[@Paul1239193]

What year does each panelist think we'll have AGI? Your answer must be a number. Should they program their AI's to, ultimately, maximize the greatest happiness function? For example, when an AGI gets a hold of all the NSA's data, what happens.

[@abdulrahmankerim2377]

We are talking about reaching Human AI. However, we cannot handle microphone problems.

[@AdeelKhan1]

46:35 Piaget has done some good work here. From what I gather. en.wikipedia.org/wiki/Jean_Piaget Personally, I am not familiar with his works.

[@bohdanafrolyak853]

fun to watch the smart people in this conversation being so polite about the idiots haha

[@twstdelf]

Yes! Love that Tom Gruber alluded to A Young Lady's Illustrated Primer!! That has been a benchmark in my mind as well.

[@fischX]

a good benchmark would be human capabilities based on human development and age

[@personzorz]

Oh no, not that crank Yudkowsky in the questions...

[@calvinsylveste8474]

Yann LeCun & Jurgen Schmidhuber in the same room, they should have made them sit next to each other :)

[@aadityaura3139]

Then generate the captions :P

[@420_gunna]

damn schmidhubby rockin the all white Also Etzioni is a sham

[@TheFrygar]

LOL - Jurgen: "How will we create AGI? We already know, we've known for 25 years. It's 2 recurrent neural networks." Uh....sure....good luck with that. I'll check back when it's been 35 years...I guess??

[@darlingtonakogo6245]

Pollen Applebee I wanted to everyone reaction after that, but it was edited out. I'd like to see Jurgen and Yoshua/Yann go back and forth on Jurgen ideas of using RNN and Metalearning to create AGI

[@DistortedV12]

This is epic

[@coffeebreakhero3743]

boring question. human level is not a scalar. Maybe instead of discussing whether it is possible to build a tower to the sky, discuss the theoretical limits to tower height.

[@Wasteomindy]

Why, one can collect a dictionary of all activities which humans has been involved in since the known history, and then check if AI is capable to perform everything from it better then best skilled in the according field human could :) I suppose that such a dictionary cannot be full, since people create new stuff to do with the time, but... you can measure the speed of filling the dictionary with new items, and when the speed drops below certain level, you can mark the dictionary a "sufficiently full list of human activities at the current datetime"

[@coffeebreakhero3743]

Nikolay Bobovnikov I agree with you. Anyway, the words Human-level are moot

[@mscottveach]

+Coffee Break Hero why do you think the words 'human-level' are moot? it's an actual practical standard that we apply all the time when evaluating machine learning performance.

[@coffeebreakhero3743]

***** alright. englighten me. define human level. after you answer, I'll reply, but what does it matter if its human level or not? if we can do better, lets do better.

[@Kelly_Jane]

Coffee Break Hero one reason is that once we reach human level ai it would be a very short time before we reach higher than human which would then help create better and better ai at an exponential rate. This is commonly referred to as the Singularity.

[@thebiggerwethinkthefasterw787]

Let's Go Team! Let's create Abundance 4 Everyone! Check out the videos on my channel and contact me if you want to support the optimistic vision of Abundance 4 Everyone. Please share what you think and feel. Let's take massive action together and help accelerate a world of abundance 4 everyone! Success is our only option guys and gals! We can do this! We will do this by working together and creating the future we want to see. Abundance 4 Everyone! Love, Love, Love. Compassion. Our Future. Looking forward to connecting with y'all.

[@9999rahul9999]

I wish you had half the number of people on this panel, like the four on the left

[@bohdanafrolyak853]

Anca so smart B-)

[@Ryan90red]

The speakers commonly shift the microphone further from themselves while talking, leaving soft gaps in their speech. Its such a small issue, but its terrible to have to hear.

[@MrAndrew535]

People refer to data acquisition like its a major accomplishment, when in fact, that is the easiest part of the scientific process. Without the ability to formulate relevant questions from which a sound understand can be derived, then regardless of the quantity of data, it is rendered useless.

[@septwolfke5204]

I cast a white stone upon the preservation and defense of innocence as well.

[@JeffBuckleyFanboy]

Oren Etzioni seems to be a very disrespectful person based upon how he acted when Mr. Schmidhuber spoke. Oren shouldn't even be there based on his limited credentials in regards to A.I.

[@AdeelKhan1]

I did not notice this. When did this happen in the video?

[@Jacen777]

Alpha Go (Deep Mind) should be introduced to the game DAY Z or ARK.

[@XOPOIIIO]

Reinforcemen learning programms are actually conscious.

[@top1percent424]

Oh c'mon you had Demis in the panel and you ask someone else to speak first.

[@KaplaBen]

Yeah it's called gallantry.

[@top1percent424]

Crouzier Benjamin what men call gallantry, and the gods adultery.

[@darlingtonakogo6245]

Top 1Percent lol Dennis is actually "booboo" on that panel, you have Yoshua Bengio, Yann Lecun and Jürgen Schmidhuber, these are Super Saiyan Gods of AI/ML...

[@brentoster]

Here is some more insight into human neuroscience and how it could be applied to an AGI: kzbin.info/www/bejne/Z5CwlKNjjs-Do7M

[@tunahelpa5433]

You'll NEVER have AGI that's based on Von Neuman machines. You must break through your preconceptions and create a NEW kind of circuit, possibly based on synthetic biology, that can do the same kind of processing that even a simple a nematode neuron can do, and then figure out a way to put 100 blion of them into a cubic foot. Good luck, but I do believe it will happen

[@Kelly_Jane]

Eliezer should have been on the panel.

[@KrishnaHarish]

Basically no one knows!

[@xyl4540]

Not having Ray on this panel was a crime.

[@trapslime16]

2221, april 14

[@bosssolidlacostranostra33]

THE NEW AMERIKA

[@Shplidaligity]

AGI could happen in 50 years or 5 years nobody has a clue.

[@JustforFun-cb7bo]

Just not possible, human is way too complex, so to make same level AI as a human's is just science fiction, i am not saying that we wont built robots with amazing AI that would handle simple jobs and tasks, BUT the process of learning is only for humans and animals (up to some point of course!) IF we make an AI that learns like we do then it's over, it will surpass us in no time.

[@Adhil_parammel]

Mark my words, ai will be biological one not technological

[@epsilon21001]

I cringed when she said I'm gonna shut up......The only female on the panel and she gets interrupted and cut short. How rude!

[@arielle675arielle]

Getting interrupted is one thing, giving the interruption full credential by saying 'ok I'm gonna shut up,' that inherently puts her message in a lower rank than the panelists, is something else. She should have never said those words. Woman or not , that's lack of self-respect.

[@rylaczero3740]

bunch of fools

[@epsilon21001]

I can't stand the swiss guy in all white...who talks like that?

[@epsilon21001]

The accent could be nice to the ear depending on the speaker. He sounds like an evil genius wannabe

[@KaplaBen]

Agreed. An evil genius wannabe that wrote the LSTM paper.

[@epsilon21001]

Jordon, thanks for pointing that out. However, to my point, I don't see this irritating attitude from the rest of the panel (which contains people who contributed even more to AI). Generally, one's contributions should be isolated from one's ego.

[@emmanuelnsanga]

He cant help but talk like that, he's not a native English speaker.

[@RandyH524]

Why are you so critical of his accent? One could possibly say the same about your accent, it's not something that you really can control it's more or less something learned.