On this day in 1903 was born John von Neumann, the greatest polymath of the 20th century. We are all using "von Neumann architecture" chips to read this tweet. Game theory, ergodic theory, Artificial Life, quantum mechanics, hydrodynamics, statistics, etc. He was an immigrant.
All human beings have incredible potential, even those who do not look or dress like us, come from somewhere else, and are poor. Be kind to everyone, not because of who they might become, or do for you, but because it is the right thing to do.
It's a deal. August 19, 2031. My prediction is that this project will not register at all in any history of the previous ten years. So we'll understand that it was a totally insignificant announcement.
August 19, 2021.
This day goes down in history as when
@elonmusk
announced Tesla Bot at Tesla AI Day.
We'll look back in 10 years and understand how significant today really was.
Let's count how many truly autonomous (no human safety driver) Tesla taxis (public chooses destination & pays) on regular streets (unrestricted human driven cars on the same streets) on December 31, 2020. It will not be a million. My prediction: zero. Count & retweet this then.
Wow!
@elonmusk
predicts autonomous taxis are imminent.
(I would love my car to earn money while I don't use it.)
It's interesting to read
@rodneyabrooks
argumentation against autonomous driving.
What will happen? I wonder..
New essay, shorter than normal. It is about how "artificial intelligence" got started 60+ years ago, and the difference between AI and AGI (spoiler alert: none). And why people mistakenly think that AGI is a "thing", with rapid progress. It's not.
Separating a child from a parent may punish a parent but it destroys a child for life. In what world is destroying a child in order to punish a parent a moral thing to do? I mostly tweet about robotics/Ai, but I will not stand by when children are destroyed... or sent to ovens.
While people are morning Opportunity remember that it was a robot built by humans and operated by humans at an average distance of 225M km (140M miles) with no chance to touch it directly, for 15 years, **sixty** times its design lifetime. Bravo humans, and what a Space Odyssey!
Hey,
@elonmusk
I see you are interviewing the guy you have endorsed for President on Aug 12th. Would be super helpful if you could get us clarity on his policy that comes up at his rallies about mass deportations. Who is going to do the round ups, and will there be any legal
My grandson is turning one year old. Compared to a six year old he is bad at manipulation, but compared to any ML/RL trained AI system he is a dexterity superstar, with a wide variety of strategies and skills he applies to novel situations. He is just a regular human kid.
1/ Mon Sept 26, 1977, 45 years ago today, I rode my bicycle up Page Mill Rd in Palo Alto, under 280, right on Arastradero, then up a driveway to a delapidated semi-circular wooden building. I walked through the front door at one end and entered the future. I have never left.
This is a really good explanation of and reminder about ChatGPT. The deep want of people (even my smart friends) to believe it is doing more is so disconcerting. There is no miracle. A new, rather tricky, tool, yes, but we're still going to have to get through life on our own.
Many people are trying to make excuses for why this trivial problem trips up LLMs. Occam's razor should remind you that there is no emergent reasoning in LLMs. Instead it gets lucky (and it is astonishing that it works so well--that says something deep about our language)
1/7 It is good to be aware of the literature. Shannon and McCarthy describe a `neural network' doing what GPT-3 does in their intro (pg vi) to Automata Studies, Princeton University Press, 1956.
Four for the price of none. I just posted my 4 part penultimate essay in my Future of AI and Robotics series. Steps Toward Super Intelligence, Parts I, II, III, & IV. Best read in order--they link to each other. Here's the first part on How We Got Here.
Here's my conversation with
@rodneyabrooks
, one of the greatest roboticists of all time, having co-founded iRobot, Rethink Robotics, Robust-AI, and before that serving as director of the MIT Computer Science and Artificial Intelligence Laboratory.
April forty three years ago my master's advisor accidentally lost my handwritten thesis draft on machine learning. There was no copy. I wrote a new version from memory and notes. It was better, but still really quite bad. Now, finally, it occurs to me: perhaps it was no accident.
Fernando Corbato of MIT has passed away. He developed Multics which begat Unix which begat Linux. 1990 Turing Award. He was my academic department head when I was a young faculty member. A very kind and wise man.
1/3 Promised to retweet this now. On April 22, 2019, Elon said there would be up to 1 million Tesla robotaxis on the road by now. I said zero. And the answer is ... zero!
Let's count how many truly autonomous (no human safety driver) Tesla taxis (public chooses destination & pays) on regular streets (unrestricted human driven cars on the same streets) on December 31, 2020. It will not be a million. My prediction: zero. Count & retweet this then.
Voyager 1, launched Sept 5, 1977, is furtherest human artifact from Earth, over 15B miles, 24B kms. Magnificent remote debugging has it back sending understandable data after 6 month outage. Flight Data Subsystem computer, 7400 series TI chips (see image), with 8K 16 bit words of
My latest
@IEEESpectrum
column. It will also appear in the October print issue. "Ask not what your AI system can do for you, but instead what it has tricked you into doing for it."
The fetish that everything should be learned from big data sets rather than developing algorithms for intelligent systems seems like Soviet Lysenkoism where it was believed that all properties of crops were absorbed from the environment. Millions of people starved to death. 1/4
Posted my fifth annual review of my dated predictions from Jan 1, 2018, about self driving cars, AI/ML & Robotics, and human spaceflight. They seemed pessimistic back then, but they are standing the test of time. BONUS five new curmudgeonly predictions.
Scientific results are now lobbied for via press releases which deliberately obfuscate the generality and robustness of the results. Then companies deny primary access or even ability to test system to people who are not obvious fan boys. Now we have politicized science. Sad.
This is a very crude and uniformed analogy. Eyes are not at all like cameras (5 orders of magnitude more dynamic range for a start), nor are neural networks anything but a crude cartoon of how brains work (which we don't know in any case...). Crude analogies cost $B's.
@teslaownersSV
@TheRealKeean
We have to solve a huge part of AI just to make cars drive themselves.
In retrospect, it was inevitable. The road system is designed for cameras (eyes) & neural nets (brains).
Next week, 40 years since I joined the tenure track CS faculty at Stanford AI Lab; had finished my PhD there in 1981. 40 years??? We all thought we would have AGI by 1983. Did that happen? No. Has it happened yet? No. In 300 years we'll look back: What were they thinking?!?
The talk about hallucinations in LLMs has gotten it all wrong. The true hallucinations are by company execs who think it is OK to release to general users products that are based on LLMs that confabulate wildly, as all LLMs do. Time will show a high price paid by society.
Patrick Winston of MIT has passed away. He was director of the MIT AI Lab for 25 years, and a professor for 50 years. The best teacher I have ever encountered, and the author and editor of many text books and compendia that defined AI. He was a gentle soul and loved by all.
Last January 1st I made predictions on self driving cars, AI and ML, and space, with dates attached, for 32 years into the future. Today I score how well these predictions stood up for the first year--only 31 more scorecards to come!
My dear and long time friend Daniel Dennett passed away today. He was a towering figure in philosophy and in particular in the philosophy of AI. Now we have only memories of him. [[And do keep up to date with your correspondence--I have a 30 day old email from him in my inbox
Hypnotism: “We’ve made a soft promise to investors that, ‘Once we build a generally intelligent system, that basically we will ask it to figure out a way to make an investment return for you.'” There are less polite words for this malfeasance.
We, humans, give way too much importance to language and symbols as the substrate of intelligence.
But primates, dogs, cats, crows, parrots, octopus, and many other animals don't have humans-like languages, yet exhibit intelligent behavior beyond that of our best AI systems.
2/N
My three laws of robotics.
The companies that I have founded have sold 50+ million real robots to real customers; research robots, home cleaning, nuclear power plant inspection, military ground robots, upper body humanoids in factories, and now at ,
AI learning performance is not yet even in the same continent as biological learning. Good to see people exploring the deep science of learning still. Readable summary: and source material:
Sometimes a few moments of deep thought ahead of spending vast sums of money can lead to insight on what the real technical difficulties might be. Just saying. But you all know that from my blog posts.
In my 1981 computer vision PhD defense John McCarthy, namer of AI, took umbrage that I did not know the cross ratio in projective geometry, saying we would talk about it in the private session--scared me almost to death. Today I learned that his PhD was on projective geometry.
SoftBank stopped manufacturing Pepper last year. Cute doesn't cut it except for very cheap robots. Robots have to provide value to customers that is more than the price of the robot. And expecting customers to find applications is a losing strategy.
We're all going to be living in the metaverse getting everything we need to know from ChatXYZ, surrounded by stable diffusion art, paid for with crypto, and delivered by self-driving cars, or flying taxis. All fusion powered and run by AGI quantum computers. So much future! Soon!
I was interviewed for this story a few weeks ago. The headline is not what I would have written, but the content about warehouse robots, and how LLMs are not the key to making them better, is what I was explaining.
Am I misremembering? Wasn't there just a little hype around how OpenAI's approach to robotics was so powerful??? "the market isn't mature enough yet"... perhaps the `problem' is that the market is mature and understands what is valuable and not.
Nils Nilsson (former Stanford CS Chair) has passed away. He lead the Shakey robot project at SRI in the 1960s, *after* his 1965 book "Learning Machines". My 1972 introduction to AI research was his 1971 book "Problem-Solving Methods in AI". Thank you Nils!!! RIP.
"Factfulness" by Hans Rosling et al, has changed my understanding of the world. It uses data to show me where and how I was wrong, and identifies 10 ways us humans consistently and wrongly misunderstand so many things. Please read it.
Very thoughtful piece on where we are with human imitative AI, and the challenges to making sensible use of data flows in our lives, by a giant in statistics and machine learning, Michael Jordan of Berkeley.
Have blogged my experience taking truly driverless Cruise rides in San Francisco. Conclusion: good progress and a solid minimal viable product. Deployable with 24hr operation in some locales. Not yet at the level that many have felt imminent for years.
Amazon Fresh failure.
My third law of robotics, shared here yesterday is:
3. Technologies for robots need 10+ years of steady improvement beyond lab demos of the target tasks to mature to low cost and to have their limitations characterized well enough that they can deliver
My annual review of my Jan 1, 2018, predictions for AI&ML, autonomous vehicles, and space. Lots of commentary on progress in 2020, and where things are slow. Also questions for AV journalists to ask AV company PR depts.
New blog post today; an essay on a somewhat esoteric topic. The history of programming languages for robot arms in factories. With stories from AI Labs at Stanford and MIT.
I often cringe when I read stories about robots, even from otherwise respectable outlets. But occasionally I start cheering when I see one that is well researched and sensible and informative. Thank you
@dalvin_brown
@washingtonpost
!! Robots in our homes.
Who are we kidding? 🙄🙄
AI is NOWHERE near automating software engineering.
Of course, a co-pilot is great for increasing productivity, but AI is strictly just an assistant to aid programming professionals.
We are at least 3-5 years away from automating software engineering.
Remember the good old days of a few months ago? Back when blockchain was about to revolutionize AI? And 3 years ago when all cars would be self driving by now? Now we have finally figured it out. Game playing AI is going to change the world! Oh wait, we did that in 1996 too...
People underestimate how long it takes to get from academic paper to real world robotics. 25 years on Disney is using my subsumption architecture for humanoid eye control, better and smoother now than our 1995 implementations on Cog and Kismet.
Back on April 1st I posted my three laws of robotics. Here are my three laws of AI.
1. When an AI system performs a task, human observers immediately estimate its general competence in areas that seem related. Usually that estimate is wildly overinflated.
2. Most successful AI
The AI hype cycle is damaging real innovation across tech. It is crazy FOMO having negative impact on the world as we move forward. "AI is artificially inflating valuation numbers across venture: The segment captured nearly 50% of all deal value in Q2. But flat and down rounds
As you know Chris, we are nowhere near AGI. Those who deploy physical machines in the real world know that. My companies have deployed millions more robots than anyone else, in five domains. I wrote theses on ML in 1977 and computer vision in 1981. It is a long game. Work hard!
Go on a podcast walk listening to brilliant people talk about how close we are to Artificial General Intelligence, return to work trying to get freakin tensorflow 2.2 to run with jetpack 4.4 just I can just get my basic toolchain to build
cc
@lexfridman
August 40 yrs ago I left my Stanford faculty position and joined the MIT faculty, still (barely) in my 20s. Had just co-founded AI software co in Palo Alto, so had $100K workstation to work on that in my $140K Lexington MA house. It had 1MB of RAM. Be restless! Be fearless!
Tele-op robots presented as autonomous. Tesla Optimus humanoid folding a shirt. 1X humanoid robots. Misrepresentations of what robots are actually doing can also be called LIES. Note that the Stanford robot cooking and cleaning videos are also tele-operated.
Oct3'18 my exporting robotics co., 250 engineers&factory workers (direct&indirect) in MA&NH was driven out of biz by retaliatory Chinese tariffs against US started trade war. I struggle with how that loss of jobs and tech leadership made America greater that what is was.
Just published my 2nd of 32 annual reviews of my January 1st 2018 hype dampening predictions about self driving cars, AI & ML, and the space industry. My predictions had dates attached so you can evaluate how well I did.
Love this paper; augmenting computation with non-Turing processes; something that most likely occurs in all living systems. This is long term research; don't think about it in terms of immediate applications; it is exploring wild new spaces.
@denizzokt
As I look at the rush to build very complex robots using new and unproven learning and control methods I am reminded of the history of combat aircraft. In the first world war competition was intense in aerial combat. Flat screen displays, though indispensable in modern fighter
Many people talking about peer review recently. Just posted a new blog post on my experiences with peer review both as an author and an editor thirty years ago.
Careful reading of AlphaZero chess paper; reasons to be cautious about the hype. In particular, to me, it looks like Stockfish (the opponent) was forced to play like a perceptual engine, which it is not, whereas AlphaZero is.
@CamachoCollados
Paul, it's a puppet. It didn't speak with anyone. It was a person using it as a puppet. 'Sophia' did not speak or 'praise'. It was a loud speaker for a person with a microphone. That is not a tech story, at least not for the last century....
People never die on FaceBook or LinkedIn. I get messages from old friends, inviting me to congratulate them on work anniversaries, or birthdays, or to become their friend, years, and now even decades, after the have passed on. The cyber world has merged with the spiritual world!
End to end learning for robotics is in a bit of a hype cycle but proponents sometimes miss the fact that practical results are not just end to end learning. As UC Berkeley Professor
@Ken_Goldberg
points out in this clip from a recent talk he gave at Stanford, Good Old Fashioned
It is now March 11 in Japan. Ten years ago today was the earthquake and tsunami which devastated four of six nuclear reactors at Fukushima Daiichi.
@iRobot
sent six robots used in the shutdown & cleanup -- still operating when I was on site in April 2014.
This age's biggest technology flub. When the flubber fails to meet the road fortunes will be unmade, jobs will lost, and investors will be hopping mad. Wishing for something to be true, even for a long time, doesn't make it so.