Hello, World!
Welcome to our 33rd edition of Hold the Code, where we explore AI’s role in customer service, its impact on students apply to which colleges, and its capabilities at the poker table.
We would also like to extend a welcome to all of our new NU subscribers who found us through our poster competition. Stay tuned for our edition on March 1st, where we will announce the winners of the ice cream prize!🍦
As always, happy reading!
When to Send in the Machines
With the advance of AI customer service agents, a new kind of question has arisen: when is it best to talk to a human? A recent paper published in the Journal of Marketing examines exactly that, looking at how customers are likely to react depending on their situation and who they’re talking to.
We’ve all been there, a streaming service isn’t working, you can’t get a refund, and there’s nothing customer service can do. This paper looked at exactly these kinds of situations and found the interesting result that we are more likely to be forgiving of unfavorable news when it is coming from an AI. Similarly, customers are more likely to come away with a positive response when we hear favorable news from an actual person.
Why is this?
Current thinking is that it all comes down to our own speculation as to why the news came out the way that it did. While in unfavorable situations there is almost always nothing the customer service agent could have done differently, when we hear the bad news from an AI, we are not plagued with doubts about underlying lack of care or selfish intentions. Similarly however, when positive news comes around, customers are likely to apply this same thinking and attribute their turn of luck not to the kindness of another human but instead to an unfeeling algorithm.
How are companies implementing this…
This raises a new opportunity for companies to prioritize which customer service roles should be replaced with AIs, sending them first to roles where negative interactions are common and reserving roles where interactions are often positive. Furthermore, companies can use this knowledge to create more human-like AIs to address positive situations, playing on customer intuition to get a more positive response from the customer in turn.
…ethically?
This knowledge raises ethical concerns though, as people worry that companies will take advantage of consumers using this methodology. That’s why it’s important for consumers to be aware of these strategies and make their decisions accordingly.
Navigating College Admissions
College is expensive. It’s a huge commitment — time-wise and money-wise — and universities depend on their student population to get them the funds they need to provide them with a quality education. But how to get those students interested in the college of their dreams, or even let them know it exists? Enter: Naviance.
What is Naviance?
If you’ve graduated from high school within the last few years, you’ve probably run into Naviance. It’s been a growing part of the way that colleges and universities are advertising themselves to potential new students, as smaller schools push for space in a highly competitive field.
The Implications of Naviance
Naviance uses a number of concerning techniques and systems to influence which students end up at certain colleges, as detailed below:
Race-based advertisements: The way that Naviance is approaching getting students and schools connected is, to say the least, concerning. Multiple colleges and universities have contracts with the company that allow them to advertise specifically to students of a certain race (even after this was supposedly removed from their advertising features). Colleges have been know to target only white students in certain states. Supposedly, this kind of targeted advertising is used as a tool to improve a school’s diversity, but if it is only targeting white students, it’s definitely not doing that…
A high school requirement: Colleges and universities know that they can reach students through Naviance because many high schools require students to spend time on the platform. These schools make their graduating seniors spend time on the portal, getting recommendation letters from teachers, and adding potential schools to their lists.
Disguised “recruitment” messages: Colleges can also send students advertisements disguised as recruitment messages. This tactic can make student believe that they are a good match for the school reaching out to them, when they may have only gotten this message as part of a marketing campaign.
“There’s some social engineering at play that feels really concerning,” said Ceceilia Parnther, a St. John’s University professor who studies higher education leadership. “I see it being an electronic form of gatekeeping.”
Evaluating Ethics
So does Naviance really provide a service to its students? Or is it having a negative effect on their perceptions of their abilities and specifically dividing them up by race? Advertising is always a complicated, multifaceted topic, especially when it comes to how companies approach their audience. Naviance, however, might be approaching their audience — a body of young adults who are trying to decide their futures — in a way that doesn’t prioritize those futures as much as they should be.
Weekly Feature: AI & Poker
In the Amazon room of Las Vegas’ Rio casino, Seth Davies has just risked it all at the Texas Hold ‘em table, having just declared himself all in for $1.7 million in chips
In his hand? Absolutely nothing.
Davies' opponent folded, and he won the round. Later that evening, he went on his computer and logged onto PioSOLVER (an AI system that determines the optimal decisions for a given poker hand). Although the system recommended that he bet a bit more on the turn, Davies felt validated to see that the $1.7 million bet on the river was the right call.
“That feels really good,” Davies recalls. “Even more than winning a huge pot. The real satisfying part is when you nail one like that.”
Stories like Davies have become increasingly common in the world of poker as AI carves out its place in this sphere. Here we recap on this NYT article that discusses the history behind poker AIs and what the implications are.
The History of AI & Poker
The pursuit of the perfect poker play dates back to 1944 with the release of Theory of Games and Economic Behavior by mathematician John von Neumann and economist Oskar Morgenstern. They explored how to quantify the decision making process and use this to derive the behaviors that humans exhibit in a social economy. To do so, they used poker as a way to model this behavior.
But why poker?
Von Neumann rejected using other games like chess or checkers as a model for his proof because in these games all of the information and decision making is visible to the players. Poker has “tactics of deception” which von Neumann thought mirrored the real world more accurately.
When to bet and when to bluff
Through this work, von Neumann found that players should bet high when they have really good hands and for a certain percentage of their really bad hands. This percentage depends on the size of the potential bet relative to the size of the pot.
Von Neumann found that when this method was used, his opponents were not able to do better than break even and — more often than not — they ended up losing.
Catching up
Von Neumann’s proof did not immediately catch on in the world of poker until Neil Burch, a computer science researcher at the University of Alberta took it to an AI company called DeepMind (which is now a subsidiary of Google). Their initial attempts to make an AI poker player were not particularly fruitful.
“If you put a knowledgeable poker player in front of a computer and let them poke at it, [the program] got crushed, absolutely smashed,” Burch said.
The initial difficulties stemmed from the large decision space that poker invites. In computer science, a structure called a decision tree is often used to represent the outcomes of a scenario or a game. These trees contain branches which each represent a single outcome. The number of branches on a tree get exponentially larger as the number of decision points increases. For example:
In the classic game Rock-Paper-Scissors, there are 9 branches total. When you play this game, you have three choices to pick (rock, paper, or scissors). Your partner also has the same three options. This means there are 3×3 (or 9) possible outcomes (you choose rock and your partner chooses rock, you choose rock and your partner chooses paper, you choose scissors and your partner chooses rock…you get the idea).
Compare this with a decision tree of 2-person Texas Hold ‘em, which has 316,000,000,000,000,000 branches for each possible outcome.
All of these outcomes were initially difficult to encode in a computer system. However as AI improved, so did its ability to model these many outcomes and learn which decisions maximized their chances of gaining the most profit.
Catching on
Once AI had improved, those in the poker world suddenly became very interested in its capabilities. One researcher from the Alberta team was paid a large sum of money to make software for poker players to teach them the best strategies to employ. Eventually, a Polish programmer and online poker player, Piotrek Lopusiewicz, released PioSOLVER, the tool used by none other than our friend Seth Davies.
Impacts on Poker
Already, the impacts of tools like PioSOLVER have been immense. The best poker players of today are able to analyze the AI’s strategies and create heuristics for themselves to use when they are dealt similar hands. Erik Seidel (a pro poker player from the 80s) said that if the players who use these AI tools went back in time just 15 years, they would crush their competition. Some in the poker world are also concerned about the possibility of cheating with the access to such tools being so high. These concerns have even driven some away from online poker competitions out of mistrust of their opponents methods and morals.
Whether or not you think AI is a good or bad addition to the poker table, it is definitely something that has changed how the game is played.
Love HTC?
Follow RAISO (our parent org) on social media for more updates, discussions, and events!
Instagram: @Raisogram
Twitter: @raisotweets
RAISO Website: https://www.raiso.org
Written by Arielle Michelman, Hope McKnight, and Molly Pribble
Edited by Molly Pribble