How Artificial intelligence and Machine Learning Can Impact Market Design

From Digital Culture & Society

Jump to: navigation, search

For millennia, markets have played a key role in providing individuals and businesses with the opportunity to gain from trade. Auctions have been commonly used to generate a setting for trade. Traditionally, markets require structure and institutional support to operate with efficiency, Market Design research however, reveals it may be critical to design auctions, and market institutions more broadly, in order to achieve efficient outcomes.

It has become increasingly difficult for market designers to know all of the information necessary to design the best institution. How could these publishers design their auctions to make the best use of their advertising space? How might they maximize returns on their activities? Can market designers learn the skills necessary to set optimal reserve prices? How can market designers efficiently learn about the environment of the market? The answer is Artificial Intelligence (AI) and Machine Learning.

AI and machine learning have emerged as important tools for market design. Data mining has become the norm for markets like eBay, Amazon, and Uber, who are identifying important patterns to help create better experiences for their customers and to increase market efficiency. AI and machine learning algorithms are helping marketplaces better anticipate consumer demand and producer supply, which help market designers better manage their environment, as well as help target products and services into segmented markets.

One interesting case of AI machine learning happens at Google. Not only is this two sided market using Artificial Intelligence to set better reserve prices and segment their consumers, they are also using it to help advertisers bid on ads. In April 2017 Google introduced its “Smart Bidding”, which utilizes AI and machine learning algorithms that use vast amounts of data to help advertisers bid automatically on advertisements based on ad conversion. This allows advertisers to determine optimal bids, and refine their own models of user conversion in order to better allocate advertising dollars. There are also many other different ways AI algorithms can be applied to market design.

AI has also played a key role in the operation of markets. Technological advancements have changed the environment vastly since the early 20th century, no longer are transportation and energy projects the norm, now the focus has shifted to a focus in communications networking.

In 2010 the Obama White House issued a National Broadband Plan, with the goal of freeing a huge amount of bandwidth from older systems, that was to be used instead as part of the modern data highway system (p. 4). The US Federal Communications Commission (FCC) designed an auction market to do part of the job.

One of the most critical algorithms used for this process was the “feasibility checker” which was developed with the aid of machine learning methods. In the FCC auction design, each bidder was quoted a price in a round-by-round auction, and decreased each round. After each round of bidding the auctions software used its feasibility checker to determine whether it could feasibly assign a station to continue broadcasting.

The performance of this descending auction design depends deeply on the quality of the feasibility checker. While many steps took place in the development of the system, there is an emphasis on the role of machine learning.

Multiple tests were conducted using 1.4 million problem instances and testing feasibility checking algorithms. Next, a automated algorithm configuration developed at the University of British Columbia was used with the idea “to start with a highly-parapeterized algorithm for solving satisfiability problems and to train a random forest model of the algorithm” and tested it to generate large datasets with parameters and performance measures (p. 11).

Next, the system identifies the parameter vector that maximizes the expected improvement in performance, given the mean and variance of the prior and the performance of the best-known parameter vector. Finally, the system tests the actual performance for the identified parameters and adds that as an observation to the dataset. Proceeding iteratively, the system identifies more parameters to test, investigates them, and adds them to the data to improve the model accuracy.

New datasets can also be created using instances where parameterized algorithms are slow (15 secs or more), and by creating new algorithms, can improve machine learning techniques and improve AI performance. Machine learning proved possible to decompose full problems into smaller ones, and to reuse old solutions as starting points for a search, and to guide solutions of further problems.

Online marketplaces like eBay have grown dramatically by providing businesses and individuals with opportunities to profit from online trading which had previously been unavailable. These online marketplaces, have made it easier for retailers to market their own goods while also moving excess inventory, while consumers enjoy the ease of searching for whatever they may want.

The effectiveness of these marketplaces aren’t without debate however. The accuracy of eBay’s popular feedback and reputation mechanism has been a point of discussion, research results show that user-generated mechanisms are often biased and suffer from grade inflation. The average positive seller percentage on eBay is about 99.4% with a median of 100%. The challenge now is to interpret the true levels of satisfaction without manipulation from sellers.

To better serve buyers these web based services collect data generated by the transactions and searches that occur on these sites daily, and leverage that data to promote trust. Two examples from recent research show how some marketplaces can apply AI to the data they generate, which can help create more trust and better experiences for their customers.

One option is using AI to assess the quality of sellers. Using Natural Language Processing (NLP), a mature area in AI, marketplaces mine data generated by messages in order to better predict the kind of features that customers value.

Another option is using AI to create a market for feedback. Feedback is a act that requires time, not all buyers will choose to leave feedback, while another problem is feedback is often inflated. Furthermore, buyers are attracted to sellers who are already established, which creates barriers for new sellers without feedback.

The Chinese marketplace Taobao implemented a clever use of NLP AI to help solve this problem, for Taobao it is the platform, that decides whether feedback is relevant and not the seller who pays for the feedback. “Importantly, feedback quality only depends on how informative it is, rather than whether the feedback is positive or negative. The AI measures the quality of feedback with a NLP algorithm that examines the message and length of feedback to finds out whether key features of the item are mentioned within the content. The roll of AI was not to reward buyers for positive feedback, but for information.

An important application of AI and machine learning in online marketplaces is the amount of interaction and engagement of buyers and their search patterns. Recent research has determined that consumers learn what they like during the search process, and that buyers search for products significantly more than studies assumed. Interestingly, searches often proceed from vague to very specific. Online marketplaces might then, design their search algorithms to understand search intent which can better serve their customers. AI and machine learning could play an important role in not only recognizing customer intent and helping that customer narrow their preferences, but segmenting customers into groups that will be better served with informative depth in their search process.

AI holds enormous potential to improve efficiency. AI can overcome computational barriers, reduce search frictions and help organizations better coordinate services. The uses of AI promise widespread benefits to all.

Personal tools
Bookmark and Share