AI driven with Analogies

We Followed Melanie Mitchell's work, Especially with regard to Analogies. A Good direction,  working with advertising we liked the idea of their use, but the approaches was not mature enough, in particular natural language methods had not emerged yet.  Would like to see where this has gone since.

The Computer Scientist Training AI to Think With Analogies   By Joe Pavlus in Quanta Mag

Melanie Mitchell has worked on digital minds for decades. She says they’ll never truly be like ours until they can make analogies.

he Pulitzer Prize-winning book Gödel, Escher, Bach inspired legions of computer scientists in 1979, but few were as inspired as Melanie Mitchell. After reading the 777-page tome, Mitchell, a high school math teacher in New York, decided she “needed to be” in artificial intelligence. She soon tracked down the book’s author, AI researcher Douglas Hofstadter, and talked him into giving her an internship. She had only taken a handful of computer science courses at the time, but he seemed impressed with her chutzpah and unconcerned about her academic credentials.

 Mitchell prepared a “last-minute” graduate school application and joined Hofstadter’s new lab at the University of Michigan in Ann Arbor. The two spent the next six years collaborating closely on Copycat,   a computer program which, in the words of its co-creators, was designed to “discover insightful analogies, and to do so in a psychologically realistic way.”

The analogies Copycat came up with were between simple patterns of letters, akin to the analogies on standardized tests. One example: “If the string ‘abc’ changes to the string ‘abd,’ what does the string ‘pqrs’ change to?” Hofstadter and Mitchell believed that understanding the cognitive process of analogy — how human beings make abstract connections between similar ideas, perceptions and experiences — would be crucial to unlocking humanlike artificial intelligence.

Mitchell maintains that analogy can go much deeper than exam-style pattern matching. “It’s understanding the essence of a situation by mapping it to another situation that is already understood,” she said. “If you tell me a story and I say, ‘Oh, the same thing happened to me,’ literally the same thing did not happen to me that happened to you, but I can make a mapping that makes it seem very analogous. It’s something that we humans do all the time without even realizing we’re doing it. We’re swimming in this sea of analogies constantly.” ...' 

Mining Scientific Applications by Analogies

This could also be done inside a company to look at publications like technical reports, and then also linking those to external publications as well.    Driven by specific goals as well.  The broad ability to use analogies usefully is a powerful thought.  The broad idea is one we looked at extensively, the tech is now here to do it.

AI analyzed 3.3 million scientific abstracts and discovered possible new materials  in MIT Technology Review

A new paper shows how natural-language processing can accelerate scientific discovery.
The context: Natural-language processing has seen major advancements in recent years, thanks to the development of unsupervised machine-learning techniques that are really good at capturing the relationships between words. They count how often and how closely words are used in relation to one another, and map those relationships in a three-dimensional vector space. The patterns can then be used to predict basic analogies like “man is to king as woman is to queen,” or to construct sentences and power things like autocomplete and other predictive text systems.

New application: A group of researchers have now used this technique to munch through 3.3 million scientific abstracts published between 1922 and 2018 in journals that would likely contain materials science research. The resulting word relationships captured fundamental knowledge within the field, including the structure of the periodic table and the way chemicals’ structures relate to their properties. The paper was published in Nature last week.

Because of the technique’s ability to compute analogies, it also found a number of chemical compounds that demonstrate properties similar to those of thermoelectric materials but have not been studied as such before. The researchers believe this could be a new way to mine existing scientific literature for previously unconsidered correlations and accelerate the advancement of research in a field.  .... "  ...

Respecting Absurd Ideas

Always liked the idea of exploring the edge of possibilities.  Here a look at how to exploit absurd suggestions.   Which break down into Respect the idea, and Play with it.   I would add,  always use the promoter of the idea as the 'expert' in its application .... they can school you in its application, or even why it won't work in context.   Stretch it.  Explore with analogies.   Absurdity can be good.

strategy+business: Corporate Strategies and News Articles on Global Business, Management, Competition and Marketing
 Two Simple Concepts for Getting the Most from Absurd Ideas ... " 

Addressing the Analogy Gap

Reminiscent of using humans as a peripheral, here determining high level relationships, then having the deep learning sort out the lower level patterns.  Points to Mechanical Turk, which we used this way.  At what point are the results general?  The comment about scale is key.  Also the mapping involved, very useful for generalizing and testing.

Crowdsourcing may have just helped close the "analogy gap" for computers    It's vexed computer scientists for decades, but a huge roadblock for true AI is falling    By Greg Nichols for Robotics

Researchers at Carnegie Mellon University (CMU) and the Hebrew University of Jerusalem in Israel have used crowdsourcing to teach computers to generate analogies so they can mine datasets to address new challenges by repurposing old concepts. "After decades of attempts, this is the first time that anyone has gained traction computationally on the analogy problem at scale," says CMU professor Aniket Kittur. The researchers hired participants via Amazon Mechanical Turk, tasking them to look through products on an innovation website and find analogous products from the same source. The participants noted which words caused them to link disparate products, mapping each pathway. Computers with deep-learning algorithms used these insights to analyze additional product descriptions and find new analogies. The researchers say this strategy can be used to customize computer programs to identify analogies between patent applications and literature on global problems.  ... " 

Learning by Sketching with CogSketch

Quite interesting idea.  I remember looking at the CogSketch platform.  Had not of thought of it as a learning method.  But many of us do learn visually, can this be a means to fill the learning gaps?   Does it work in contexts other than geology, that are less visual?   How do the analogy aspect work?

Helping Students Learn by Sketching
Sketch Worksheets software analyzes and provides feedback on student sketches

Northwestern University professor Ken Forbus and his team have developed Sketch Worksheets, software that helps students learn via sketching exercises and also provides on-the-spot feedback by analyzing sketches and comparing them to the instructor's drawings. The software is founded on CogSketch, an artificial intelligence platform previously developed in Forbus' lab that employs visual-processing algorithms to automatically replicate and understand human-drawn sketches. 

Sketch Worksheets' comparisons of student and instructor sketches are conducted by an analogy model, in which students and instructors apply conceptual labels to their sketches to represent relationships among the drawings' different components. Forbus says CogSketch uses analogy to compare labels and give feedback. Geoscientists at the University of Wisconsin-Madison used Sketch Worksheets to devise a set of 26 sketches that cover topics in introductory classes. "This is a step in creating software that can communicate with people as flexibly as we communicate with each other," Forbus says. .... " 

See the tag below for more research information regarding Ken Forbus, previously mentioned.

Analogous Search Leading to Intelligence?

Have always been interested in how analogy can be used, and how it can be used to convince.   Much used in advertising and marketing. Certainly something we do as part of what we call intelligence. Now think if this as a way of manipulating Google's Knowledge Graph.  Look at 'found in related search'.   In Google Operating System.  Makes me think about alternate kinds of search among stored concepts.  How might these include analogy relationships?

Learning by Analogy

Particularly interesting suggestion of how people are 'intelligent'.

Making Computers Reason and Learn by Analogy

Structure-mapping engine enables computers to reason and learn like humans, including solving moral dilemmas    by Amanda Morris
 
Northwestern Engineering’s Ken Forbus is closing the gap between humans and machines.

Using cognitive science theories, Forbus and his collaborators have developed a model that could give computers the ability to reason more like humans and even make moral decisions. Called the structure-mapping engine (SME), the new model is capable of analogical problem solving, including capturing the way humans spontaneously use analogies between situations to solve moral dilemmas.

“In terms of thinking like humans, analogies are where it’s at,” said Forbus, Walter P. Murphy Professor of Electrical Engineering and Computer Science in Northwestern’s McCormick School of Engineering. “Humans use relational statements fluidly to describe things, solve problems, indicate causality, and weigh moral dilemmas.” ... ' 

See more on Structure Mapping Engines (SME).

Learning with Less Data

In Technology Review: Learning things with far fewer examples.    Pushes at the edges of what is meant by learning,  and varying kinds of errors that define success in varying contexts.    Also, at the edge of innovative creativity and analogy.

New Science Search Engine Searches for Meaning

We always wanted better semantic search.  Based on semantic meaning,  underlying scientific ontology,  common sense,  business driven analogy,  even goal based inference.   Is this what this new AI based approach is?   A quick few searches found it to be interesting.  Mostly brought down abstracts,  not including corpus.   Found no analogy evident, but evidence of that could take some effort.  Comparison to 'Google Scholar', which I recall claimed some semantic network?

" ... Artificial-intelligence institute launches free science search engine
Semantic Scholar comes from centre backed by Microsoft co-founder Paul Allen.  ... Called Semantic Scholar.  ... " 

Nicola Jones     In Nature,

Further covered in Technology Review,

Biologically Inspired Design

Attended a talk today in the Cog Sci Institute on the status of work being done at Ga Tech on Biologically Inspired design, or Biomimicry.  Given by Prof. Ashok Goel    Most often examples in this space look at how designs in nature can be used to address specific human engineering problems.  Most of what they are doing today is trying to understand how human knowledge about biology can be retrieved to address engineering requirements.  The problem is that engineers and biologists don't speak the same technical language.  Also, we need to understand how to search and reason by analogy in both domains.  The extension of analogy is a form of creativity.  We researched reasoning by analogy when addressing consumer needs.   This can then be linked to cognition, as part of Human centered computing. All this is a crucial part of AI.  There is much more work to be done.

Talk slides: https://www.slideshare.net/secret/9S42InuTABn0qY

Taking Autocompletion Further

A simplistic metaphor for intelligence is autocompletion.  We are very used to it today. We have knowledge, now how do we use that knowledge to predict what should be next in a given context?    In Wired:

" .... Now, a government-backed research team wants to provide similar suggestions to the world’s programmers as they’re writing computer code. That’s right: the aim is to guess what programmers are coding before they code it.

This week, Rice University said that Darpa, the Pentagon’s mad science division, has invested $11 million in this autocomplete programming project, dubbed PLINY, after the ancient Roman author of the first encyclopedia, “Text search prediction is the best analogy,” says Vivek Sarkar, the chair of the computer science department at Rice and the principal investigator on the project. “People will be able to will be able to pick from a list of possible solutions.” ... " 

Network as Nervous System

A reasonable analogy for digital enterprise.    And Analytics should be the brain, placing the network n contexts to gather as much useful data as possible.  " ... Business and network strategies will become more integrated as the network becomes the nervous system of the digital business ... "  

Product Development Using Structured Analogies

Just received this from SAS.  Note that it is an ad and you have to provide information to get the background paper, have not done that yet.  Click here for more.  I am intrigued because we experimented with a similar approach a decade ago, and analogies are powerful ways to generate new product ideas.  May be something useful here.

" ... New Product Forecasting Using Structured Analogies

Learn about a new patent-pending approach that may be helpful in certain new product forecasting situations. Make manual overrides to the statistical forecasts, and get a better sense of the risks and uncertainties in new product forecasts through visualization of past new product introductions. ...  They describe it further:

SAS has a new patent-pending approach to NPF that combines the use of analogies with structured judgment. This "structured analogy" process for new product forecasting has six main steps:

Query step: Find a set of candidate products that have similar attributes to the new product.

Filter step: Manually remove inappropriate or outlier products from the set of candidate products.

Cluster step: Cluster the candidate products according to their sales pattern, and manually select the most appropriate cluster to serve as the surrogate products.

Model step: Select the most appropriate statistical model for the cluster of surrogate products, and extract the statistical model features.

Forecast step: Use the extracted statistical model features to forecast the new product. ... " 

Finding Things at Lowe's

Not a new thing.  Using an App to find things in a complex space like Retail.  We examined this, starting with kiosk systems.  But I like the implication that things can be found by association with related things. Associating things on the shelf with needs?  Or even by analogy,  Or with a Watson AI?