Artificial intelligence (AI) is proving very adept at certain tasks – like inventing human faces that don’t actually exist, or winning games of poker – but these networks still struggle when it comes to something humans do naturally: imagine.-

Once human beings know what a cat is, we can easily imagine a cat of a different color, or a cat in a different pose, or a cat in different surroundings. For AI networks, that’s much harder, even though they can recognize a cat when they see it (with enough training).

To try and unlock AI’s capacity for imagination, researchers have come up with a new method for enabling artificial intelligence systems to work out what an object should look like, even if they’ve never actually seen one exactly like it before.

“We were inspired by human visual generalization capabilities to try to simulate human imagination in machines,” says computer scientist Yunhao Ge from the University of Southern California (USC).-

“Humans can separate their learned knowledge by attributes – for instance, shape, pose, position, color – and then recombine them to imagine a new object. Our paper attempts to simulate this process using neural networks.”

A new “red team” will try to anticipate and thwart attacks on machine learning programs.

THE PENTAGON SEES  artificial intelligence as a way to outfox, outmaneuver, and dominate future adversaries. But the brittle nature of AI means that without due care, the technology could perhaps hand enemies a new way to attack.

The Joint Artificial Intelligence Center, created by the Pentagon to help the US military make use of AI, recently formed a unit to collect, vet, and distribute open source and industry machine learning models to groups across the Department of Defense. Part of that effort points to a key challenge with using AI for military ends. A machine learning “red team,” known as the Test and Evaluation Group, will probe pretrained models for weaknesses. Another cybersecurity team examines AI code and data for hidden vulnerabilities.

Machine learning, the technique behind modern AI, represents a fundamentally different, often more powerful, way to write computer code. Instead of writing rules for a machine to follow, machine learning generates its own rules by learning from data. The trouble is, this learning process, along with artifacts or errors in the training data, can cause AI models to behave in strange or unpredictable ways.

“For some applications, machine learning software is just a bajillion times better than traditional software,” says Gregory Allen, director of strategy and policy at the JAIC. But, he adds, machine learning “also breaks in different ways than traditional software.”

By Poornima Apte

There’s Netflix for movies. Stitch Fix for clothes. Hungryroot, an AI-powered delivery service, hopes to occupy a similar niche for online groceries in the United States.

The recommender system uses a collaborative filtering, supervised learning model to match consumer preferences to foods. Customers answer questions about their dietary habits, the kinds of foods they (and family members) like, the family size, budget, and more. On a weekly basis, the Hungryroot algorithm predicts the groceries the customer might like. Once the customer approves the list, a box ships from one of three Hungryroot locations. Customers also receive a set of recipes, also predicted by the algorithm, that use the week’s ingredients.

Neil Saunders, the managing director of GlobalData’s retail division, has seen grocery retailers of all stripes lean into AI as a way of better forecasting demand. “With the disruption from the pandemic and more people buying groceries online, demand forecasting has become increasingly difficult for retailers and AI can help them make sense of the data and make more accurate decisions about what to stock,” Saunders says.