To the extent that the Mitch Hedberg quote above is funny, the humor resides in the multiple meanings of the initial sentence “I used to do drugs”. The more obvious meaning of that sentence is a statement about the past where the phrasing suggests that the present is drug free. The bottom part of the meme shows a drug-filled present, and we reread the original and see the other meaning as equally grammatical.

The humor resides in understanding both meanings simultaneously. If a person sees only one meaning, no humor. But equally importantly, if they read the later part and…


The Bottle

John was strolling down the beach when he spotted a glass bottle washed ashore. Curious, he picked it up and saw that it was empty but for a piece of paper with some typed message. When he looked closer at the message, he was disappointed. It wasn’t in English. It looked Russian. Luckily, he had a Russian-speaking friend. “It is not Russian,” the friend said, “but Ukrainian. I think I can manage to make sense of it”. …


The goals of this post are modest. The previous post about chess thinking outlined how humans don’t first generate a list of legal moves that they filter down. It hand-wavingly claimed that moves “pop up” in our minds. This post shows one possible mechanism of such popping up.

The analysis will be based on what I call “situational categories” (or situations, for short), a term I prefer over the overused term pattern. I hope to sidestep the rigid imagery that the term pattern evokes: it makes some people think of assorted pieces arrayed in some specific configuration. By contrast, the…


Chess offers us an interesting window into cognition. Millions of chess games are available for free. Games come labeled with the rough level of expertise (available from rank novices through super-grandmasters) and also the available amount of time (ranging from just three minutes for all moves in bullet chess through a leisurely week or longer per move in correspondence chess). Furthermore, given the strong chess engines today that are magnitudes better than the best human players, it is possible to automatically label blunders (although we cannot accurately identify small mistakes).

In this post, I dwell on one aspect of chess…


Word meaning is fluid and shaped by expectations. Just as a fluid takes the shape of its container, so too the context of a word forms a container shaping the word meaning. To take a simple example, consider a word you know well, “Starbucks”, and look at its meaning in these two sentences: “I worked at Starbucks last year” and “I drank my Starbucks”.

In the first sentence, “I worked at __” leaves a place-of-work shaped hole and in the second “I drank ___” leaves a drink shaped hole, and the word Starbucks happily morphs into either shape quite effortlessly…


It is a commonplace notion that most words have a particular meaning (with the caveat that sometimes they have a handful of meanings for ambiguous words such as jaguar). Understanding a meaning of a word in a sentence, according to this default theory, is one of meaning selection, not one of meaning creation. Here I argue that this simple theory is insufficient, and creativity is required in understanding any sentence.

Dictionaries lead us to believe that a word has a certain set of meanings. The word pitcher, my 11th-edition Webster Collegiate Dictionary tells me, has three meanings: (A) a person…


What symbols are for Marcus differs from what symbols are for Bengio — and in between those extremes exists another possibility that offers a compromise.

My research lives midway between the two extremes. I want symbols, but I don’t wish them handcrafted: my work has been in obtaining symbols by statistical techniques by reading from tons of data (from web pages while at Google, and from text and graphs at Pinterest). My symbols have a distributed representation and graded evocation, and I am after compositionality and higher level structures. Finally, one key ambition of the cognitive architecture I built for…


PatiencePhoto by Michel Porro on Unsplash

Why is AI Brittle?

What ails AI today is brittleness. Someone unveils this fantastic AI program that shows superhuman performance on some task, but a minor tweak to the input brings it down to its knees. AI systems that are marvelous at recognizing objects in images can be duped by altering a single pixel into seeing a non-existent giraffe. Self-driving cars, which scream “AI!!!” more loudly than all else, are hoodwinked by mere stickers into driving on the wrong side of the road.

What makes AI brittle? I present one overlooked reason that I consider key to this brittleness: a heavy reliance on eager…


I suggest they don’t.

Frederick A. Cook’s picture of Ed Barrill atop a peak claimed to be Denali but actually 15,000 ft lower.

This post may be read by itself or as the second installment of “The Emperor’s New Benchmarks”. That post looked at a benchmark of passing curiosity (pun detection), but now we consider a problem that commands the community’s ongoing veneration.

Recognizing Textual Entailment

Several NLP benchmarks test for semantic understanding. One such is Recognizing Textual Entailment (RTE): Given two sentences, a premise P and a hypothesis H, decide if we can conclude H given P. This comes in two flavors. For binary RTE, the answer may only be yes or no. …


The Dangers of NLP Benchmark Oversimplification

Language is complex. For NLP, we often use simplified benchmarks, relegating phenomena such as metaphor, metonymy, and ungrammatical usage to future work. We also simplify in other ways. Surely an innocuous step — doesn’t science proceed from the simple to the complex?

This post documents damage simplistic benchmarks cause. While I had grand designs for this post, I will scale down to one particular benchmark — pun detection — and follow up with other posts for other benchmarks.

Methodology: To illustrate the dangers I perceive, I use published concrete examples from reputable venues. Singling out papers is not my style…

Abhijit Mahabal

I do unsupervised concept discovery at Pinterest (and previously at Google). Twitter: @amahabal

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store