Making AI systems that see the world as humans do

Making AI systems that see the world as humans do
An example question from the Raven's Progressive Matrices standardized test. The test taker should choose answer D because the relationships between it and the other elements in the bottom row are most similar to the relationships between the elements of the top rows. Credit: Ken Forbus

A Northwestern University team developed a new computational model that performs at human levels on a standard intelligence test. This work is an important step toward making artificial intelligence systems that see and understand the world as humans do.

"The model performs in the 75th percentile for American adults, making it better than average," said Northwestern Engineering's Ken Forbus. "The problems that are hard for people are also hard for the model, providing additional evidence that its operation is capturing some important properties of human cognition."

The new computational model is built on CogSketch, an platform previously developed in Forbus' laboratory. The platform has the ability to solve visual problems and understand sketches in order to give immediate, interactive feedback. CogSketch also incorporates a computational model of analogy, based on Northwestern psychology professor Dedre Gentner's structure-mapping theory. (Gentner received the 2016 David E. Rumelhart Prize for her work on this theory.)

Forbus, Walter P. Murphy Professor of Electrical Engineering and Computer Science at Northwestern's McCormick School of Engineering, developed the model with Andrew Lovett, a former Northwestern postdoctoral researcher in psychology. Their research was published online this month in the journal Psychological Review.

The ability to solve complex visual problems is one of the hallmarks of human intelligence. Developing that have this ability not only provides new evidence for the importance of symbolic representations and analogy in visual reasoning, but it could potentially shrink the gap between computer and .

While Forbus and Lovett's system can be used to model general visual problem-solving phenomena, they specifically tested it on Raven's Progressive Matrices, a nonverbal standardized test that measures abstract reasoning. All of the test's problems consist of a matrix with one image missing. The test taker is given six to eight choices with which to best complete the matrix. Forbus and Lovett's performed better than the average American.

"The Raven's test is the best existing predictor of what psychologists call 'fluid intelligence, or the general ability to think abstractly, reason, identify patterns, solve problems, and discern relationships,'" said Lovett, now a researcher at the US Naval Research Laboratory. "Our results suggest that the ability to flexibly use relational representations, comparing and reinterpreting them, is important for fluid intelligence."

The ability to use and understand sophisticated relational representations is a key to higher-order cognition. Relational representations connect entities and ideas such as "the clock is above the door" or "pressure differences cause water to flow." These types of comparisons are crucial for making and understanding analogies, which humans use to solve problems, weigh moral dilemmas, and describe the world around them.

"Most artificial intelligence research today concerning vision focuses on recognition, or labeling what is in a scene rather than reasoning about it," Forbus said. "But recognition is only useful if it supports subsequent reasoning. Our research provides an important step toward understanding visual reasoning more broadly."


Explore further

Structure-mapping engine enables computers to reason and learn like humans, including solving moral dilemmas

More information: Andrew Lovett et al. Modeling visual problem solving as analogical reasoning., Psychological Review (2017). DOI: 10.1037/rev0000039
Journal information: Psychological Review

Citation: Making AI systems that see the world as humans do (2017, January 19) retrieved 25 June 2019 from https://phys.org/news/2017-01-ai-world-humans.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.
698 shares

Feedback to editors

User comments

Jan 19, 2017
they specifically tested it on Raven's Progressive Matrices


The raven progressive matrices have the problem of being culture specific, with hidden assumptions such as clockwise rotation as default, or the assumption that a figure is handled as a geometric figure instead of an abstract symbol.

When you take that into account, many of the matrices become ambiguous with multiple right answers, and the real test of intelligence would be to explain why you chose the pattern instead of just picking one out.

In the example matrix, if we don't assume the arrows and triangles to be the same geometric object rotated, but abstract symbols with different meanings when the arrow/triangle points left and right, then the right answer can't be D because that would be the same symbol on both sides. That would make a syntax of A->B, A->B, A->A and where's the pattern in that?

Maybe the matrix is a kind of rosetta stone. Obviously, the last symbol must be G because it's different

Jan 19, 2017
If you're not convinced, imagine that we assing the arrow symbols actual semantic values, "Left", "top", "right", and imagine that this is someone's attempt to explain that to us by placing the symbols over an object. Maybe they're trying to convey a language to someone who doesn't already understand it.

So if the bottom left rectangle means "left", and the question is what is the missing symbol for "right", why would it be the same rectangle? Therefore D must be wrong, and the only possible answer is G - the information we get is that there exists this irregularity that has something to do with the shape or meaning of the constant object in each row - kinda like an irregular verb or noun.

In this way, one can reason alternate answers to the Raven matrices that are just as plausible as the "correct" answer. That means instead of testing intelligence, they test whether the test taker assumes the same facts as the person who made the test.

Jan 19, 2017
If you're not convinced, imagine that we assing the arrow symbols actual semantic values, "Left", "top", "right", and imagine that this is someone's attempt to explain that to us by placing the symbols over an object. Maybe they're trying to convey a language to someone who doesn't already understand it.

So if the bottom left rectangle means "left", and the question is what is the missing symbol for "right", why would it be the same rectangle? Therefore D must be wrong, and the only possible answer is G - the information we get is that there exists this irregularity that has something to do with the shape or meaning of the constant object in each row - kinda like an irregular verb or noun.

In this way, one can reason alternate answers to the Raven matrices that are just as plausible as the "correct" answer. That means instead of testing intelligence, they test whether the test taker assumes the same facts as the person who made the test.

I would have said "A" or "I"...

Jan 19, 2017
Not so much a random number generator but fuzzy probabilistic non-integer logic. Which enables an AI to recognise "probably a cat" as it has been trained on many other cats.

Probabilistic additive logic also helps to parallelise the object matching processing algorithm. Just like our brains are parallel not serial computers and are much faster because of this despite our slow clock rate. Some processor offloading will be required to handle the parallel additive function of estimating probabilities: As you cant have 99.9% of your numerous processors all waiting for a byte/word in a locked data block to become free, or invalidating all their cached data block every time the data is added/subtracted and unlocked by another single process.

With the massive increase in interest in AI (from self drive cars to customer service) 2017 should be a very interesting year for new computer architectures:- https://www.nextplatform.com

Jan 19, 2017
Making AI systems that see the world as humans do
Why? So they can act as translators and liaisons like C3PO? Machines need to see the world like machines do.

Like any animal, humans see the world in terms of surviving to reproduce.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more