The research in the area of machine learning and AI, now a key technology in virtually every industry and business, is far too voluminous for anyone to read it all. This column aims to bring together some of the most relevant recent findings and papers – particularly, but not limited to, artificial intelligence – and explain why they matter.
This week in AI, scientists conducted a fascinating experiment to predict how “market-driven” platforms like food delivery and ride-sharing companies affect the overall economy when optimized for different purposes, like revenue maximization. Elsewhere, demonstrating the versatility of AI, a team from ETH Zurich developed a system that can read tree height from satellite images, while a separate group of researchers tested a system to predict the height of trees. startup success from public web data.
The market-driven platform’s work leverages Salesforce’s AI Economist, an open-source research environment for understanding how AI could improve economic policy. In fact, some of the researchers behind the AI Economist were involved in the new work, which was detailed in a study originally published in March.
As the co-authors explained to TechCrunch via email, the goal was to investigate two-sided marketplaces like Amazon, DoorDash, Uber, and TaskRabbit that enjoy greater market power due to the increase in demand and supply. Using reinforcement learning — a type of AI system that learns to solve a multilevel problem through trial and error — researchers trained a system to understand the impact of interactions between platforms (e.g., Lyft ) and consumers (e.g. passengers).
“We use reinforcement learning to reason about how a platform would perform under different design goals… [Our] The simulator allows evaluation of reinforcement learning policies in various contexts under different objectives and model assumptions,” the co-authors told TechCrunch via email. “We explored a total of 15 different market parameters – i.e. a combination of market structure, buyers’ knowledge of sellers, [economic] shock intensity and design objective.
Using their AI system, the researchers came to the conclusion that a platform designed to maximize revenue tends to increase fees and extract more profits from buyers and sellers during economic shocks at the expense of well-being. be social. When platform fees are fixed (for example, due to regulation), they found that the incentive to maximize a platform’s revenue typically aligns with user welfare considerations. global economy.
The results may not be earth-shattering, but the co-authors believe the system – which they plan to open – could provide a basis for a business or policymaker to analyze a platform economy under different conditions, designs and regulatory considerations. “We adopt reinforcement learning as a methodology to describe the strategic operations of platform firms that optimize their pricing and matching in response to changes in the environment, be it economic shock or regulation,” they added. . “It may yield new insights into platform economies that go beyond this work or those that can be generated analytically.”
Turning our attention from platform companies to the venture capital that fuels them, researchers from Skopai, a startup that uses AI to characterize companies based on criteria such as technology, market and financials, claim to be able to predict a startup’s ability to attract investment using publicly available data. Drawing on data from startup websites, social media, and business registries, the co-authors say they can achieve prediction results “comparable to those also using structured data available in databases. private”.
The application of AI to due diligence is nothing new. Correlation Ventures, EQT Ventures and Signalfire are among the companies currently using algorithms to inform their investments. Gartner predicts that 75% of venture capitalists will use AI to make investment decisions by 2025, up from less than 5% today. But while some see the value in the technology, dangers lurk below the surface. In 2020, Harvard Business Review (HBR) found that an investment algorithm outperformed novice investors but had biases, such as frequently selecting white, male entrepreneurs. HBR noted that this mirrors the real world, pointing to AI’s tendency to amplify existing biases.
In more encouraging news, scientists from MIT, alongside researchers from Cornell and Microsoft, claim to have developed a computer vision algorithm – STEGO – that can identify images down to the individual pixel. While this may not sound significant, it is a vast improvement over the conventional method of “teaching” an algorithm to spot and classify objects in images and video.
Traditionally, computer vision algorithms learn to recognize objects (eg, trees, cars, tumors, etc.) by showing them many examples of objects that have been labeled by humans. STEGO removes this time-consuming and laborious workflow by instead applying a class label to each pixel in the image. The system isn’t perfect – it sometimes confuses oatmeal with pasta, for example – but STEGO can successfully segment things like roads, people and traffic signs, the researchers say.
When it comes to object recognition, it looks like we are approaching the day when academic works like DALL-E 2, OpenAI’s image generation system, will be produced. New research from Columbia University shows a system called Opal that is designed to create featured images for news stories from text descriptions, guiding users through the process with visual prompts.
When tested with a group of users, researchers said those who tried Opal were “more effective” at creating featured images for articles, creating more than twice as many “usable” results. that users without. It’s not hard to imagine a tool like Opal eventually making its way into content management systems like WordPress, perhaps as a plugin or extension.
“Given the text of an article, Opal guides users through a structured search for visual concepts and provides pipelines for users to illustrate based on an article’s tone, topics, and illustration style” , wrote the co-authors. “[Opal] generates various sets of editorial illustrations, graphic assets and concept ideas. »