Cloud GPU providers are sitting at the center of several megatrends: the AI and machine learning boom, the commoditization of GPU-aware apps, and the overall push to rent rather than buy expensive computing resources. My analysis of these trends and why GPU clouds have come of age for SiliconANGLE here.
Artificial intelligence — its value, risks and utility in enterprise scenarios — not surprisingly dominated the discussion at this week’s MIT CIO Symposium, one of the year’s biggest gatherings of senior information technology executives. In this post for SiliconANGLE, Paul Gillin and I review what some of the CIO panelists revealed about the state of their domains, and their relationship with AI tools.
The U.K.’s National Cyber Security Center last week posted a joint blog with the nation’s regulatory commissioner’s office about the need for better cybersecurity breach transparency. They’re concerned about the unreported incidents, in particular ransomware cases, which are getting more dangerous, more prevalent and more costly. The situation creates a vicious cycle: “If attacks are covered up, the criminals enjoy greater success, and more attacks take place,” they wrote in the post.
In this analysis for SiliconANGLE, I look at the implications for designing the next generation of customer support systems using AI enhanced tools.
Most of us have been interacting with customer support agents for years. It can be a frustrating experience: Oftentimes the agent knows less than we do about their product or service, calls are dropped or transferred to other agents. About two years ago, I had such a bad experience with AT&T Inc.’s customer support that I ended up cancelling my cell and internet service with the company.
But now there are artificial intelligence chatbots and chat programs that are supposed to make our lives better. With all the attention focused on ChatGPT and other AI-based chatbots, a new long-term research study has found that AI can help improve support, but only under carefully controlled situations. Let’s examine the specific circumstances and what’s in store for the future of support. In this post for SiliconANGLE, I dive into what they found and make some recommendations on how to be more effective at deploying AI for customer support situations.
All this chatter about ChatGPT and large language models interests me, but from a slightly different perspective. You see, back in those pre-PC days when I was in grad school at Stanford, I was building mathematical models as part of getting my degree in Operations Research. You might not be familiar with this degree, but basically was applying various mathematical techniques to solving real-world problems. OR got its beginnings trying to find German submarines and aircraft in WWII, and then got popular for all sorts of industrial and commercial applications post-war. As a newly minted math undergrad, the field had a lot of appeal, and at its heart was building the right model.
Model building may bring up all sorts of memories of putting together plastic replicas of cars and ships and planes that one would buy in hobby stores. But the math models were a lot less tangible and required some careful thinking about the equations you chose and what assumptions you made, especially on the initial data set that you would to train the model.
Does this sound familiar? Yes, but then and now couldn’t be more different.
For my class, I recall the problems that we had to solve each week weren’t easy. One week we had to build a model to figure out which school in Palo Alto we would recommend closing, given declining enrollment across the district, a very touchy subject then and now. Another week we were looking at revising the standards for breast cancer screening: at what age and under what circumstances do you make these recommendations? These problems could take tens of hours to come up with a working (or not) model.
I spoke with Adam Borison, a former Stanford Engineering colleague who was involved in my math modeling class: “The problems we were addressing in the 1970s were dealing with novel situations, and figuring out what to do, rather than what we had to know built around judgment, not around data. Tasks like forecasting natural gas prices. There was a lot of talk about how to structure and draw conclusions from Bayesian belief nets which pre-dated the computing era. These techniques have been around for decades, but the big difference with today’s models is the huge increment in computing power and storage capacity that we have available. That is why today’s models are more data heavy, taking advantage of heuristics.”
Things started to change in the 1990s when Microsoft Excel introduced its “Solver” feature, which allowed you to run linear programming models. This was a big step, because prior to this we had to write the code ourselves, which was a painful and specialized process, and the basic foundation of my grad school classes. (On the Stanford OR faculty when I was there were George Danzig and Gerald Lieberman, the two guys that invented the basic techniques.) My first LP models were written on punched cards, which made them difficult to debug and change. A single typo in our code would prevent our programs from running. Once Excel became the basic building block of modeling, we had other tools such as Tableau that were designed from the ground up for data analysis and visualizations. This was another big step, because sometimes the visualizations showed flaws in your model, or suggested different approaches.
Another step forward with modeling was the era of big data, and one example with the Kaggle data science contests. These have been around for more than a decade and did a lot to stimulate interest in the modeling field. Participants are challenged to build models for a variety of commercial and social causes, such as working on Parkinson’s cures. Call it the gamification of modeling, something that was unthinkable back in the 1970s.
But now we have the rise of the chatbots, which have put math models front and center, for good and for bad. Borison and I are both somewhat hesitant about these kinds of models, because they aren’t necessarily about the analysis of data. Back in my Stanford days, we could fit all of our training data on a single sheet of paper, and that was probably being generous. With cloud storage, you can have a gazillion bytes of data that a model can consume in a matter of milliseconds, but trying to get a feel for that amount of data is tough to do. “Even using ChatGPT, you still have to develop engineering principles for your model,” says Borison.”And that is a very hard problem. The chatbots seem particularly well-suited to the modern fast fail ethos, where a modeler tries to quickly simulate something, and then revise it many times.” But that doesn’t mean that you have to be good at analysis, just making educated guesses or crafting the right prompts. Having a class in the “art of chatbot prompt crafting” doesn’t quite have the same ring to it.
Who knows where we will end up with these latest models? It is certainly a far cry from finding the German subs in the North Atlantic, or optimizing the shortest path for a delivery route, or the other things that OR folks did back in the day.
When we first thought about the plausible future of a real Skynet, many of us assumed it would take the form of a mainframe or room-sized computer that would be firing death rays to eliminate us puny humans. But now the concept has taken a much more insidious form as — a chatbot?
Don’t laugh. It could happen. AI-based chatbots have gotten so good, they are being used in clever ways: to write poems, songs, and TV scripts, to answering trivia questions and even writing computer code. An earlier version was great at penning Twitter-ready misinformation.
The latest version is called ChatGPT which is created by OpenAI and based on its autocomplete text generator GPT-3.5. One author turned it loose on trying to write a story pitch.Yikes!
The first skirmish happened recently over at Stack Overflow, a website that is used by coders to find answers to common programming problems. Trouble is, ChatGPT’s answers are so good that they at first blush seem right, but upon further analysis, they are wrong. Conspiracy theories abound. But for now, Stack Overflow has banned the bot from its forums. “ChatGPT makes it too easy for users to generate responses and flood the site with answers that seem correct at first glance but are often wrong on close examination,” according to this post over on The Verge. The site has been flooded by thousands of bot-generated answers, making it difficult for moderators to sift through them.
It may be time to welcome our new AI-based overlords.
There is an increasing number of AI-based tools that are being used in the hiring and HR process. I am not sure whether this is a benefit to job seekers and to the employment business. Certainly, there are plenty of horror stories, such as this selection from 2020’s most significant AI-based failures such as deepfake bots, biased predictions of pre-criminal intent, and so forth. (And this study by Pew is also worth reading.)
I would argue that AI has more of a PR than HR problem, with the mother lode being traced back to the Terminator movies and Minority Report, with Asimov’s Three Laws of Robotics thrown in for good measure. In a post that I did for Avast’s blog last fall, I examined some of the ethical and bias issues around AI. Part of the issue is that we still need to encode human judgment into some digital form. And people aren’t as consistent as machines — which sometimes is a useful thing. I will give you an example at the end of this post.
But let’s examine what is going on with HR-related AI. In a study done last year by HRExaminer, identified a dozen hiring-based AI tools, with half of them focusing on the recruiting function. I would urge you to examine this list and see if any of them are being used at your workplace, or as part of your own job search and hiring process.
One of the ones on the list is HiredScore, which offers an all-purpose HR solution using various AI methods to rank potential job candidates, recommend internal employees for open positions, and measure inclusion and diversity. That is a lot of places where the doomsday “Skynet” scenario of the machines taking over could happen, and is probably one of the few plot lines that Philip K. Dick never anticipated. Still, the company claims they have trained their machine learning algorithms with more than 25M resumes and twice as many job postings.
There are other niche products, such as Xref’s online reference checking or the testing prowess of TestGorilla. The latter offers a library of more than 135 “scientifically validated tests” for job-specific skills like coding or digital marketing, as well as more general skills like critical thinking. That one struck another nerve for me. The reason I put that phrase in quotes is because I can’t validate its claim.
As many of you who have followed my work have found out, my first job in publishing was working for PC Week when it was part of the Ziff Davis corporation. ZD had a rule that required every potential hire to submit to a personality test before getting a job offer. I have no recollection of the actual test questions all these years later, but obviously I passed and so began my writing career.
In the modern era, we now have vendors that use AI tools to help screen applicants. I am not sure I would have passed these tests if ZD had them available back in the day. That doesn’t make me feel better about using AI to help assist in this process.
Let me give you a final example. When I went to visit my daughter last month, I was given a specific time period that I was allowed to enter Israel. Only it wasn’t specific: the approval was granted for “two weeks” but not starting from any specific time of day. I interpreted it one way. The German gate agents at Frankfort interpreted another way. Ultimately, the Israeli authorities at the airport agreed with my point of view and let me board my final flight. If a machine had screened me, I would have probably not been allowed to enter Israel.
In my post for Avast’s blog last year, I mention several issues surrounding bias: in the diversity of the programming team creating the algorithms, in understanding the difference between causation and correlation, and in interpreting the implied ethical standards of the actual algorithms themselves. These are all tricky issues, and made even more so when you are deciding on the fate of potential job applicants. Proceed with caution.
I tried to remove my own data recently and found it to be a very frustrating online rabbit hole. You will find either task to be nearly impossible and, sadly, this is by intent and by design: They charge by the gigabyte and aren’t paid for being accurate. And you don’t pay them anything, so you aren’t really the customer; you are just the unwilling victim.
Note: these brokers are the legitimate side of selling your data, and not to be confused with the dark web illegal side, such as the recent scraping of 700M LinkedIn users. FIghting that is for another post.
I started out my own quest by submitting removal requests for my data to three places: Epsilon, Experian, and Intelius. I picked these somewhat at random, but the trio gives you a good idea of what you are in for. My journey through this looking glass is chronicled for my latest blog post for Avast here.
In today’s changing times, tech companies must renew their focus on customers, and use their data effectively to create a holistic, 360-degree view of those customers. With this view in place, they can both improve the customer experience and better inform product development in order to attract new customers and retain existing customers. Facing fragmented data, slow and fragile data pipelines, growing demands and increasing costs, legacy data warehouse solutions are no longer sufficient. Enter next gen Cloud Data Platforms. With integrated data and seamless sharing, tech companies can now serve real-time analytics, scale up operations, and enhance the customer experience. This will take you to the slide deck for an IDG webinar that I did for Snowflake.
A startup here in St. Louis is trying to marry the analytics of the web with the practice of addiction counseling and psychotherapy. In doing so, they are trying to bring the methods of Moneyball to improve therapeutic outcomes. It is an interesting idea, to be sure.
The firm is called Takoda, and it is the work of several people: David Patterson Silver Wolf, an academic researcher; Ken Zheng, their business manager; Josh Fischer, their co-founder and CTO; and Jake Webb, their web developer. I spoke to Fischer who works full time for Bayer, and supports Takoda on his own time as they bootstrap the venture. “It is hard to put all the various pieces together in a single company, which is probably why no one else has tried to do this before,” he told me recently.
The idea is to measure therapists based on patient performance during treatment, just like Moneyball measured runs delivered by each baseball player as their performance measurement. But unlike baseball, there is no single metric that everyone has created, certainly not as obvious as RBIs or homers.
We are at a unique time in the healthcare industrial complex today. Everyone has multiple electronic health records that are stored in vast digital coffins; so named because this is where data usually goes to die. Even if we see mostly doctors in a single practice group, chances are our electronic medical records are stored in various data silos all over the place, without the ability to link them together in any meaningful fashion.
On top of this, the vast majority of therapists have their own paper-based data coffins: file cabinets full of treatment notes that are rarely consulted again. Takoda is trying to open these repositories, without breaching any patient data privacy or HIPAA regulations.
Part of the problem is that when someone seeks treatment, they don’t necessary learn how to get better or move beyond their addiction issues while they are in their therapist’s office. They have to do this on their own time, interacting with their families and friends, in their own communities and environment.
Another part of the problem is in how we select a therapist to see for the first time. Often, we get a personal referral, or else we hear about a particular office practice. When we walk in the door, we are usually assigned a therapist based on who is “up” – meaning the next person who has the lightest caseload or who is free at that particular moment when a patient walks in the door. This is how many retail sales operations work. The sole design criterion was to evenly distribute leads and potential customers. That is a bad idea and I will get to why in a moment.
Finally, the therapy industry uses two modalities that tend to make success difficult. One is that “good enough” is acceptable, rather than pursuing true excellence or curing a patient’s problem. When we seek medical care for something physically wrong with us, we can find the best surgeon, the best cardiologist, the best whatever. We look at their education, their experience, and so forth. Patients don’t have any way to do this when they seek counseling. The other issue is that therapists aren’t necessarily rewarded for excellence, and often practices let a lot of mediocre treatment slide. Both aren’t optimal, to be sure.
So along comes Takoda, who is trying to change how care is delivered, how success is measured, and whether we can match the right therapists to the patients to have the best treatment outcomes. That is a tall order, to be sure.
Takoda put together its analytics software and began building its product about a year ago. First they thought they could create something that is an add-on to the electronic health systems already in use, but quickly realized that wasn’t going to be possible. They decided to work with a local clinic here. The clinic agreed to be a proving ground for the technology and see if their methods work. They picked this clinic for geographic convenience (since the principals of the firm are also here in St. Louis) and because they already see numerous patients who are motivated to try to resolve their addiction issues. Also, the clinic accepts insurance payments. (Many therapists don’t deal with insurers at all.) They wanted insurers involved because many of them are moving in the direction of paying for therapy only if the provider can measure and show patient progress. While many insurers will pay for treatment, regardless of result, that is evolving. Finally, the company recognized that opioid abuse has slammed the therapy world, making treatment more difficult and challenging existing practices, so the industry is ripe for a change. Takoda recognizes that this is a niche market, but they had to start somewhere. “So we are going to reinvent this industry from the ground up,” said Fischer.
So what does their system do? First off, it uses research to better match patients with therapists, rather than leave this to chance or the “ups” system that has been used for decades. Research has shown that matching gender and race between the two can help or hurt treatment outcomes, using very rough success measures.
Second, it builds in some pretty clever stuff, such as using your smartphone to create geofences around potentially risky locations for each individual patient, and providing a warning signal to encourage the patient to steer clear of these locations.
Finally, their system will “allow practice offices to see how their therapists are performing and look carefully at the demographics,” said Fischer. “We have to change the dynamic of how therapy care is being done and how therapists are rated, to better inform patients.”
It is too early to tell if Takoda will succeed or not, but if they do, the potential benefits are clear. Just like in Moneyball, where a poorly-performing team won more games, they hope to see a transformation in the therapy world with a lot more patient “wins” too.