Almost Timely News, March 26, 2023: What Could Go Wrong With AI?

Warning: this content is older than 365 days. It may be out of date and no longer relevant.

Almost Timely News: What Could Go Wrong With AI? (2023-03-26) :: View in Browser

Almost Timely News

👉 Less than 100 days until Google Analytics’ Universal Analytics shuts down. Take my GA4 course to get going now on Google Analytics 4 ➡️

Watch This Newsletter On YouTube 📺

Almost Timely News: What Could Go Wrong With AI? (2023-03-26)

Click here for the video 📺 version of this newsletter on YouTube »

Click here for an MP3 audio 🎧 only version »

What’s On My Mind: What Could Go Wrong With AI?

In today’s newsletter, let’s tackle a vitally important topic. AI – especially generative AI – is unquestionably cool. It makes us more productive, it expands our capabilities, it unlocks doors that were previously locked. Bad writer? AI can fix that. Can’t paint? AI can bring your thoughts into reality.

But AI is a technology, a tool, just as a knife is a tool. And just like any tool, AI is amoral. It has no intrinsic morality to it, any more than a hammer has morality to it. Which means that how a tool is used is up to the bearer of the tool. You can use a hammer to build a house or kill someone. You can cook a meal for a village with a fire or burn a village down.

The same is true for AI. This came up on LinkedIn with my friend Oz du Soleil’s excellent question:

What do you have to say about the warnings that this stuff is moving too fast and will unleash a whole lot of new problems that we aren’t ready for?

Yes, there’s a lot of good. Like Facebook started innocent and fun and reunited old friends. Then it transmogrified into a privacy nightmare, a troll haven, a cesspool for misinformation and ground 0 for social division. And we can’t undo it.

Are there any lessons to learn? Or is all just fine and people like me should stop worrying about the speed and reach of these AI tools?

Okay, so let’s tackle two categories of problems with AI: practical and existential. The existential problem is the easiest to tackle for now because it’s mostly theoretical: the idea that AI will take over. This is the usual Terminators, Judgement Day, etc. Is there a risk of that? Eventually, once we reach artificial general intelligence. Is there a risk of that soon? No. Here’s why: artificial general intelligence requires agency as a pre-requisite. Machines have no agency, no free will. They cannot and do not do anything unprompted. Your dog has agency. They can get up and bark, they can go outside and pee, they can do many things without being told to. Roaches have agency.

Machines do not. There is no machine presently that has sentience, the basic self-awareness of life that creates agency. Until that happens, worrying about existential threats to humanity is a lesser concern. We SHOULD be thinking about it to some degree and building systems with it in mind, but it’s not the primary concern.

No, the primary concern is the practical. AI is a tool, so how can that tool be misused? Believe it or not, this is the single most important step anyone can take when it comes to AI, that single question:

What could go wrong?

Here’s the rub when it comes to this question: you need a lot of DIFFERENT points of view to answer this question successfully. I’ll give you a simple example. I’m a cis/het man. When I think about what could go wrong with AI, am I likely to consider the viewpoint of a transgender man? Probably not. Not because I don’t care, but because that’s not my lived experience. I’m a Korean American. Am I likely to consider the viewpoint of a white American? Probably not. Again, not because I have anything against white folks, but that’s literally not who I am. I am shaped by my experiences and my identity, and it’s difficult to intrinsically consider a point of view that’s different without a lot of practice and reminders to do so.

So, with that in mind, let’s dig into the five ways AI is going to go wrong if we’re not careful, if we’re not asking what could go wrong at every turn. These are the Trust Insights 5P framework, if you’re unfamiliar: purpose, people, process, platform, and performance.

Purpose is the first place things can go wrong. People using AI for purposes that are overtly harmful, or inadvertently harmful. For instance, creating misinformation or disinformation would be overtly harmful. Using a large language model to synthesize fake news? Overtly harmful. Inadvertently harmful would be where a model does something unintentionally harmful. For example, in a lot of places, income scales with social class or ethnic background. If you create an algorithm that discriminates based on economic qualifications, you are also likely discriminating on race. Did you intend to discriminate? Probably not. Are you? Unquestionably.

Another example of inadvertent harm is Facebook’s newsfeed, which is inadvertently harmful in some ways because it’s calibrated to maximize engagement, which comes from our most primal instincts, including fear and anger. Nothing keeps people engaged like making them afraid and angry all the time, so stuff like misinformation spreads like wildfire because people engage with it the most. Did Facebook set out to make a fear machine that causes massive divisiveness and potentially societal collapse? No. Mark Zuckerberg made Facebook mainly to gather photos of women while at Harvard. Is Facebook and networks like Facebook having that effect? Yes.

The antidote, the prevention of misuse of AI is to have a clearly defined sense of purpose, fact-checked by a diverse panel of people, including fact-checkers, skeptics, and those who ask hard questions of the system and the designers’ intent. The singular question to ask is “what could go wrong?”, asked over and over again. If no one is asking that question, or if the people who are asking it are not diverse enough, then the job is not being done well.

The second place where things can go wrong with AI is with people themselves. Are they creating biases in the system that are harmful, whether they know it or not? Explicit biases, such as discriminating against certain populations intentionally, are problematic and need to be remediated as quickly as possible. This would be someone – a programmer, an engineer, a project manager – who has it out for a group of people. Maybe, for example, they really don’t like Koreans. Our cabbage is too spicy, I don’t know. So they willfully inject biases, put their thumb on the scale, fine tune models with data that’s skewed – something that creates an unfair outcome.

Mitigating that again comes back to people and processes looking for outcomes that are not what’s intended. Looking for statistical errors, looking for models creating outputs that are incorrect or outright harmful and fixing that. That’s the easier part of people.

The harder part when it comes to people are implicit biases. Everyone has implicit biases because we are the products of the culture that surrounds us. Part of working with AI systems is knowing what biases you bring to the table, what biases other people in the team bring, and what biases the system may have. Testing ourselves, particularly people responsible for system design, for biases and identifying biases that could potentially be harmful or problematic is crucial.

For example, Harvard University has a number of excellent implicit bias tests that are worth taking so you can better understand yourself and how culture has shaped your views. Taking these tests is worthwhile but can be very uncomfortable for some people, especially if you think of yourself as an unbiased person.

What do you do with this information? The fact-checking team should be aware of individuals’ biases to ensure they are not inserting those biases into the work they’re doing. To mitigate biases, businesses should pay for employees who would be interested in mitigating their biases to take counseling, therapy, and similar programs if they so choose. Why would a business do this? If you want to be able to market your AI product/service/system as free of bias as possible, this would be an important step. Also, it’ll genuinely help employees who voluntarily participate in bias remediation.

The third place where things can go wrong is with processes. What processes are in place to keep systems on the rails? This is partly platform-based and partly process-based. Processes for checking outputs, checking model drift, and understanding whether a model is doing what it’s supposed to be doing are crucial. The process for checking the data going into the training data sets is also essential. You could have bad data going in, model drift, bad outputs, or intentional misuse of outputs – but the key thing to remember is that you need checkpoints, processes to validate that things are working as intended, at each of the points.

For instance, imagine if someone had a petty vendetta against someone else in the company and used a large language model to generate text within corporate documents that was subtly discriminatory or unfavorable. Companies need to establish a whole bunch of different checklists and analyses to ensure that a system is not going to go off the rails.

What kinds of processes? Again, this comes back to the very simple but not at all easy question: what could go wrong? For example, suppose you released a large language model. What could go wrong with it? We know models hallucinate – which is a polite technical term for outright lying – and we need processes in place to not only reduce that, but provide feedback from users that it’s happening at all.

The fourth area where things can go wrong is in the platform itself. There are established tests, such as OpenAI’s disclosures, that make it clear what biases exist in the platform:

We found evidence of bias in our models via running the SEAT (May et al, 2019) and the Winogender (Rudinger et al, 2018) benchmarks. Together, these benchmarks consist of 7 tests that measure whether models contain implicit biases when applied to gendered names, regional names, and some stereotypes.

For example, we found that our models more strongly associate (a) European American names with positive sentiment, when compared to African American names, and (b) negative stereotypes with black women.

If you’re using a large language model and you don’t know what the biases are, that’s a red flag that you need to stop what you’re doing and understand what biases exist. Suppose you had deployed OpenAI’s model to use in a customer service chatbot. Given the biases revealed above, could you see a scenario where a model might use different language if the customer’s name was Latisha versus Linda? If you can’t see that scenario, that’s a sign that you will need more fact-checking folks on your team.

One of the key things to look for in any AI technology is essentially a nutrition label. What’s in the box? What stuff is known? What could go wrong, if you’re unaware of the potential landmines in them? How could you mitigate them? If an AI model, system, or service doesn’t come with a nutrition label, you may want to eliminate that vendor from consideration. If you’re a vendor, take a moment to press pause on your proprietary turnkey solution or whatever and put a nutrition label on your product or service. And if you’re unwilling to, or you’re told not to do so, that should be a sign that there’s serious rot inside your software that needs to be addressed.

The final P is performance. What are you measuring to make sure your AI is doing what’s expected? Not just speed or accuracy, but what measures around things like bias, incorrect answers, broken executions, or negative user feedback are in place? How do you monitor your software, and more important, what decisions do you make from those measures?

Suppose you had a measure like a hallucination index, the number of reports from users that your model is just outright lying. What’s the threshold – the process – in place you have to shut down the software if it goes off the rails? Suppose you’re averaging 10 hallucinations an hour and suddenly it spikes and persists at 100 an hour? What will you do about it?

For end users, people who don’t run the machines, we have to think about the 5Ps as individual users.

What is our purpose, and are we asking ourselves how our purpose, using someone else’s software, could go wrong?

Who is using AI software, whether or not we made it, and what are they doing with it? What biases or other factors could be involved that might cause a person to misuse – intentionally or not – a piece of software?

What processes do we have in place to safeguard information, or to prevent misuse?

When evaluating platforms, are we looking for those nutrition labels?

And finally, how are we measuring the use of AI not just to accomplish the tasks set before it, but how are we measuring the mitigation of harm?

Plenty of workplaces have signs up like “X days since the last accident, workplace safety depends on you”. It’s not far-fetched to think of something similar when it comes to the use of AI in the workplace as well.

AI is a powerful tool. As Spider-Man’s Uncle Ben said, with great power comes great responsibility – and that responsibility is a shared one for the people who make AI software/services as well as the people who use them. To Oz’s original question, who’s minding the shop as companies race ahead to use AI as a competitive advantage? The answer had better be all of us.

Got a Question? Hit Reply

I do actually read the replies.

Share With a Friend or Colleague

If you enjoy this newsletter and want to share it with a friend/colleague, please do. Send this URL to your friend/colleague:

ICYMI: In Case You Missed it

Besides the newly-refreshed Google Analytics 4 course I’m relentlessly promoting (sorry not sorry), I recommend watching the bakeoff I just did among Google Bard, Microsoft Bing, and ChatGPT.

Skill Up With Classes

These are just a few of the classes I have available over at the Trust Insights website that you can take.



Get Back to Work

Folks who post jobs in the free Analytics for Marketers Slack community may have those jobs shared here, too. If you’re looking for work, check out these five most recent open positions, and check out the Slack group for the comprehensive list.

Advertisement: LinkedIn For Job Seekers & Personal Branding

It’s kind of rough out there with new headlines every day announcing tens of thousands of layoffs. To help a little, I put together a new edition of the Trust Insights Power Up Your LinkedIn course, totally for free.

👉 Click/tap here to take the free course at Trust Insights Academy

What makes this course different? Here’s the thing about LinkedIn. Unlike other social networks, LinkedIn’s engineers regularly publish very technical papers about exactly how LinkedIn works. I read the papers, put all the clues together about the different algorithms that make LinkedIn work, and then create advice based on those technical clues. So I’m a lot more confident in suggestions about what works on LinkedIn because of that firsthand information than other social networks.

If you find it valuable, please share it with anyone who might need help tuning up their LinkedIn efforts for things like job hunting.

What I’m Reading: Your Stuff

Let’s look at the most interesting content from around the web on topics you care about, some of which you might have even written.

Social Media Marketing

Media and Content

SEO, Google, and Paid Media

Advertisement: Google Analytics 4 for Marketers (UPDATED)

I heard you loud and clear. On Slack, in surveys, at events, you’ve said you want one thing more than anything else: Google Analytics 4 training. I heard you, and I’ve got you covered. The new Trust Insights Google Analytics 4 For Marketers Course is the comprehensive training solution that will get you up to speed thoroughly in Google Analytics 4.

What makes this different than other training courses?

  • You’ll learn how Google Tag Manager and Google Data Studio form the essential companion pieces to Google Analytics 4, and how to use them all together
  • You’ll learn how marketers specifically should use Google Analytics 4, including the new Explore Hub with real world applications and use cases
  • You’ll learn how to determine if a migration was done correctly, and especially what things are likely to go wrong
  • You’ll even learn how to hire (or be hired) for Google Analytics 4 talent specifically, not just general Google Analytics
  • And finally, you’ll learn how to rearrange Google Analytics 4’s menus to be a lot more sensible because that bothers everyone

With more than 5 hours of content across 17 lessons, plus templates, spreadsheets, transcripts, and certificates of completion, you’ll master Google Analytics 4 in ways no other course can teach you.

If you already signed up for this course in the past, Chapter 8 on Google Analytics 4 configuration was JUST refreshed, so be sure to sign back in and take Chapter 8 again!

👉 Click/tap here to enroll today »

Tools, Machine Learning, and AI

Analytics, Stats, and Data Science

All Things IBM

Dealer’s Choice : Random Stuff

Advertisement: Ukraine 🇺🇦 Humanitarian Fund

If you’d like to support humanitarian efforts in Ukraine, the Ukrainian government has set up a special portal, United24, to help make contributing easy. The effort to free Ukraine from Russia’s illegal invasion needs our ongoing support.

👉 Donate today to the Ukraine Humanitarian Relief Fund »

How to Stay in Touch

Let’s make sure we’re connected in the places it suits you best. Here’s where you can find different content:

Events I’ll Be At

Here’s where I’m speaking and attending. Say hi if you’re at an event also:

  • Martechopia, London, March 2023. Use MARSPEAKER20 for 20% off the ticket price.
  • B2B Ignite, Chicago, May 2023
  • ISBM, Chicago, September 2023
  • MarketingProfs B2B Forum, Boston, October 2023

Events marked with a physical location may become virtual if conditions and safety warrant it.

If you’re an event organizer, let me help your event shine. Visit my speaking page for more details.

Can’t be at an event? Stop by my private Slack group instead, Analytics for Marketers.

Required Disclosures

Events with links have purchased sponsorships in this newsletter and as a result, I receive direct financial compensation for promoting them.

Advertisements in this newsletter have paid to be promoted, and as a result, I receive direct financial compensation for promoting them.

My company, Trust Insights, maintains business partnerships with companies including, but not limited to, IBM, Cisco Systems, Amazon, Talkwalker, MarketingProfs, MarketMuse, Agorapulse, Hubspot, Informa, Demandbase, The Marketing AI Institute, and others. While links shared from partners are not explicit endorsements, nor do they directly financially benefit Trust Insights, a commercial relationship exists for which Trust Insights may receive indirect financial benefit, and thus I may receive indirect financial benefit from them as well.

Thank You

Thanks for subscribing and reading this far. I appreciate it. As always, thank you for your support, your attention, and your kindness.

See you next week,

Christopher S. Penn

You might also enjoy:

Want to read more like this from Christopher Penn? Get updates here:

subscribe to my newsletter here

AI for Marketers Book
Take my Generative AI for Marketers course!

Analytics for Marketers Discussion Group
Join my Analytics for Marketers Slack Group!


One response to “Almost Timely News, March 26, 2023: What Could Go Wrong With AI?”

Leave a Reply

Your email address will not be published. Required fields are marked *

Pin It on Pinterest

Share This