24,944 karma · joined September 26, 2015
What is reasoning?
In computer science and AI when we say "reasoning" we mean that we have a theory and we can derive the consequences of the theory by application of some inference procedure.
A theory is a set of facts and rules about some environment of interest: the real world, mathematics, language, etc. Facts are things we know (or assume) to be true: they can be direct observations, or implied, guesses. Rules are conditionally true and so most easily understood as implications: if we know some facts are true we can conclude that some other facts must also be true. An inference procedure is some system of rules, separate from the theory, that tells us how we can combine the rules and facts of the theory to squeeze out new facts, or new rules.
There are three types of reasoning, what we may call modes of inference: deduction, induction and abduction. Informally, deduction means that we start with a set of rules and derive new unobserved facts, implied by the rules; induction means that we start with a set of rules and some observations and derive new rules that imply the observations; and abduction means that we start with some rules and some observations and derive new unobserved facts that imply the observations.
It's easier to understand all this with examples.
One example of deductive reasoning is planning, or automated planning and scheduling, a field of classical AI research. Planning is the "model-based approach to autonomous behaviour", according to the textbook on planning by Geffner and Bonnet. An autonomous agent starts with a "model" that describes the environment in which the agent is to operate as a set of entities with discrete states, and a set of actions that the agent can take to change those states. The agent is given a goal, an instance of its model, and it must find a sequence of actions, that we call a "plan", to take the entities in the model from their current state to the state in the goal. This is usually achieved by casting the planning problem as pathfinding over a graph with a search algorithm like A*. Here, the agent's model is a theory, the search algorithm is the inference procedure, and the plan is a consequence of the theory. Deductive reasoning can be sound, as long as the facts and rules in the theory are correct: from correct premises we can deduce correct conclusions. We know of sound deductive inference rules, e.g. A*, and Resolution, used in automated theorem proving and SAT-Solving, are sound.
The classic example of inductive reasoning is inferring the colour of swans. Most swans are white (apparently) so if we have only seen white swans we have no reason to believe there are any other colours: we are forced to infer that all swans are white. We may only be disabused of our fallacy if we happen to observe a swan that is not white, e.g. a black swan. But who is to say when such a magnificent creature will grace us with its presence, outside of Tchaikovsky's ballets? Induction is thus revealed to be unsound: even given true premises we can still arrive at the wrong conclusions. Another example is the scientific method: imagine an idealised scientist, perfectly spherical, in a frictionless vacuum. She starts with a scientific theory, then goes out into the world and makes new observations about a phenomenon not described by her theory. She constructs a hypothesis to extend her theory so as to explain the new observations. The hypothesis is a set of rules, where the premises are the consequences of the rules in her initial theory. Then, being an idealised scientist, she goes looking for new observations to refute her hypothesis. Science only gives us the tools to know when we're wrong.
Abductive reasoning is the mode of inference exemplified by Sherlock Holmes. We can imagine Sherlock and Watson standing outside a tavern in London, watching as a gentleman of interest steps out of the tavern with egg on his lapel. "Ah, my dear Watson, what can we conclude from this observation?". "Why my dear Holmes, we can conclude that the man had eggs for breakfast". Holmes and Watson can arrive at this conclusion, about a fact that they have not directly observed, because they have a theory with a rule that says "if one eats eggs, one may get some on one's lapels". Working backwards from this rule, and their observation of egg on the man's lapels, they can guess that he had eggs even if they didn't directly observe him doing so. Abduction is also unsound: the man may have swapped coats with an accomplice, who was the one who had eggs for breakfast instead.
And now you know what "reasoning" means. So the next time someone asks: "what is reasoning?", you can let them know and turn the discussion to more interesting, more productive directions.
The Bitter Lesson Rich Sutton March 13, 2019
http://www.incompleteideas.net/IncIdeas/BitterLesson.html
And even fewer are aware of the author's follow up on what his article says about the current trend in AI:
Silicon Valley Doesn't Understand The Bitter Lesson – Richard Sutton
As usual with robot tech demos: WYSIWYG.
It's difficult to grok this because if you watch a human folding a t-shirt, you can reliably predict that the same human will fold a different t-shirt just as well, and in fact be perfectly capable of folding a wide variety of other clothes items as well. Not so for robots. With robots, what you see is precisely what you get. If you see a robot folding a t-shirt, all that means is that that particular robot can fold that particular t-shirt. The state of the art today is that the same robot cannot be expected to be able to fold a different t-shirt.
For example, see this article about Mobile ALOHA at Google. There's a passage where the visiting, awe-struck, journalist asks whether the robot he's just seen tying up a pair of shoelaces can tie up his own shoe.
“If I gave it my shoe,” I ventured, “would it just totally fail?”
“We could try,” Tompson said. I removed my right sneaker, with apologies to anyone forced to handle it. Tompson gamely placed it on the table, while Driess reloaded the policy.
“To set expectations,” Driess said, “this is a task that is thought of as being impossible.”
Tompson eyed his new experimental subject with some trepidation. “Very short shoelaces,” he said.
The policy booted up, and the claws set to work. This time, they poked at the shoelace without getting a grip. “Do you give consent for your shoe to be destroyed?” Driess joked, as the hands grabbed at the tongue. Tompson let them try for a few more seconds before hitting the Failure pedal.
https://archive.ph/CiJJG#selection-1887.0-1915.291
(Original: https://www.newyorker.com/magazine/2024/12/02/a-revolution-i...)
As to ACT-2 which basically uses the same techniques as ALOHA (imitation learning) far as I can tell, that's a commercial product and the information they give on their site is difficult to parse. E.g. they say they have 99.1% ±0.3 success rate, 778 successful folds and 9 garment types which is low enough to engender some trust they're not trying to inflate their numbers, but they don't say whether they trained on the garments used in evaluation or not. Chances are they did, because that's the current limit of the technology, i.e. if the garment being folded is unseen (as opposed to the environment, which they tout) then performance is basically random. So either ACT-2 have a major breakthrough that is a few leaps and bounds away from the current state of the art, or you've just watched a tech demo.
The fact that they only advertise "9 garment types" though is a big hint: they have the same problem with generalisation as everybody else at this point in time.
Then there's the problem of generalisation. RL is insane in figuring out the dynamics of any arbitrary environment you drop it in at any time. Unfortunately once you have a policy trained in one environment (or one task) you have to train again for the next environment (or task) you want to deal with. The number of different environments, situations or tasks in the real world ends up being overwhelmingly large.
So for example DeepMind has been trying for ever to train various robots to do stuff like grasp arbitrary objects by training in the real world or in simulation and still grasping is an unsolved problem, in the general case; and in fact in most cases.
With LLMs it was different in that there was a www of text to train on and the stakes are lower because there's no sim-2-real gap and there's no need to deal with the real world, in training or deployment, it's just text on a screen. That's very hard to replicate in robotics.
There are major breakthroughs still to be made and they aren't currently even on the horizon. And no, I don't have the answer with my super secrete robotic AGI homebrew project either, just saying :P
The consensus is that artists' colours are hazardous waste that needs safe disposal. You can find instructions how to do that in many arts supplies' sellers sites and the like, for example one I've used:
https://www.jacksonsart.com/blog/wp-content/uploads/2025/04/...
The comment about not licking your brushes is a reference to the common practice of licking them in the wargaming miniatures painting community (you know, Warhammer and all that). You can find videos of influencers explicitly telling you that the best way to shape the tip of your brush or remove excess water is to lick it. For example:
https://youtu.be/BXMhwPAee4U?si=WM-FXuxDtToNpkm4&t=129
If you ask for a rationale it's usually that wargaming and generally hobbyist paints are non-toxic, but the truth is that hobby paint manufacturers never list their paints' ingredients so there is no way to know. And of course there's different degrees of toxicity, not everything needs to be cobalt(II) oxide-level toxic to hurt you if you consume "large enough" quantities of it; the question is what that "large enough" means and without knowing the substance, there's no way to know.
So: don't lick your brushes.
I've sure given and taken informal feedback in the past from people I know and that's great but peer review also makes the whole process more formal and less sloppy; at least sometimes and depending on the venue and to my experience.
So basically leave it up to the Area Chair. It still helps them a bit that you give the paper a quick once over and flag it up if it's somehow complete crap. The problem is that I'm probably missing the chance to review papers I'd be really interested in and that I could help to improve.
I guess it's a bit ridiculous that AI conferences in particular can't find a systematic way to route papers to subject matter experts. I guess the usual system with area and sub-area keywords is overwhelmed by the extreme increase in numbers. This year AAAI made link to my Google Scholar and DBLP, link my top five papers, and add some more info about my subject to my profile; and they still failed to rate the papers that were really relevant as most relevant. At least the bidding pool was a bit smaller this time so I guess they're trying to work something out.
Meanwhile submission numbers are exploding and I fear all the conferences can do is try to play catch up.
Plagiarism brings a slap on the wrist unless the case is very high profile, like the example posted by azan_20 in the sibling where the plagiarising academic had many accolades and a big reputation. If the target or source of plagiarism is not famous it's like stealing bikes.
And of course if you're a senior academic being accused of misconduct by a less senior academic you can LOL about it because it's all a big joke and the less senior academic is the one who'll get a hit to their reputation for making a fuss and rocking the boat.
https://leanprover.zulipchat.com/#narrow/channel/270676-lean...
https://jmlr.org/editorial-board.html
Scroll down where it says "JMLR Editorial board of reviewers" and where you can find the names of reviewers and their areas of expetise.
The advice you get from such reviewers, even if your paper is rejected, is an invaluable tool to help you improve your work and its presentation. You eventually learn to not be sore about it and iterate until your work gets accepted.
Of course nobody's forcing you to publish, especially in a journal (in AI and CS conference proceedings tend to have much quicker turn around times and you naturally get more feedback if you get to present at a conference) but the alternative is to sit alone in your room hacking away at your code until you discover the perpetual motion machine.
Gee, I wonder what was wrong with the previous blue pigments and why it was so important to have this one under an exclusive license.
Cobalt blue is a blue pigment made by sintering cobalt(II) oxide with aluminium(III) oxide (alumina) at 1200 °C. Chemically, cobalt blue pigment is cobalt(II) oxide-aluminium oxide, or cobalt(II) aluminate, CoAl2O4. Cobalt blue is lighter and less intense than the (iron-cyanide based) pigment Prussian blue.
https://en.wikipedia.org/wiki/Cobalt_blue
Oh right.
P.S. Don't lick your brushes.
There's already plenty of systems like that, e.g. Ukraine's TW 12.7 [1] or even the MAUL used in rescue missions [2].
What is the niche targeted by Unitree with this robot? What are they selling, exactly; and to whom?
It's worth asking those questions if one wants to go beyond the excitement of watching a cool demo, and towards understanding the state of the art in robotics. Lots of cool videos of robots doing all sorts of impressive stuff around, but those are heavily edited and carefully curated to show what the vendor wants to show. The question is how much of that ends up in use in real-world situations, from homes, to roads to battlefields. The answer usually is: very little.
_____________________
[1] https://theconversation.com/ukraines-killer-robots-show-how-...
[2] https://www.telegraph.co.uk/world-news/2025/11/17/robot-coff...
If it were possible to make legged robots with the same fine motor control and dexterity as human legs, let alone four-legged animals' legs, we'd be way ahead of where we are right now in mobile robotics. We can't.
And then a tiny little one on the bum like a cutesy tail <3
>> High Payload:
>> Static 150 kg;
>> Continuous walking payload approx. 16 kg
And on top of that:
>> Long Endurance:
>> Unloaded 3h+ (>33 km);
>> Loaded > 2h (>16 km)
16 kgs for two hours max is not "extremely valuable to any military". It's hardly even valuable to a leisure camper. You might be able to get this to carry some of your stuff if you go on a short hike, but if you were thinking of going camping and having it carry your camping gear, you can forget about it.
There's a video where it's going uphill over uneven terrain with a load at a much slower pace, which is the same as what Big Dog could do 21 years ago, and another one where it's shown rolling up a smooth, grassy slope while loaded, which is pretty meh.
And that's just carrying a load. Now, you want it to carry an assault rifle and fire it? Like, at enemy targets? And hit them?
You will need to wait for a couple of major breakthroughs in robotics and AI still, I'm afraid.
5 Euro - La Callas
10 Euro - Beethoven
20 Euro - Marie Curie
50 Euro - Cervantes
100 Euro - Da Vinci
200 Euro - ???
I'm sure I know who the 200 Euro person is but I can't name her.
I'm sad I can't recognise any of the birds with any certainty at all. Fortunately their names are listed in Design B. In fact, so are some of their characteristics like size, weight, wingspan etc, so maybe you can use the banknotes to play some kind of bird card game of sorts.
Can you post some links to examples of such companies? It sounds really bad but I don't want to really believe it until I can see it with mine own eyes.
It would be nice to have concrete examples of that kind of project.
Besides which you have no idea whether that's how the Jacobian conjecture counterexample was verified. Nobody does because no details were released.