Content Authenticity Statement
100% of this newsletter was written by me, the human. When AI is used to generate content in this newsletter, I’ll prominently disclose it.
Watch This Newsletter On YouTube 📺
What’s On My Mind: Hot Takes on AI Congressional Hearing
I took a few hours to read the Congressional testimony of Sam Altman, Gary Marcus, and Christina Montgomery in depth, which you can find here. It’s always challenging reading Congressional testimony, mainly because senators spend so much time posturing in their questions that half the transcript is usually a waste. Maybe I need to write a prompt that distills down senatorial questions to just their raw question and preserves the answers of witnesses in full to make the transcripts more bearable.
Anyway, I have some in-depth thoughts about the way some AI leaders and elected officials are thinking about AI and… I’m not super encouraged. First, a few folks have asked why AI business leaders are calling for regulation. It’s not entirely altruistic; some of the suggestions like a regulatory body would inherently benefit those companies with deep pockets to be able to comply with regulations, squeezing out smaller competitors. As with all things governmental and political, any time someone’s lobbying for something, you should spend some time thinking about what’s in it for them.
On the topic of regulating models, that ship has largely sailed. With so many excellent and ever-evolving open source models, the door to regulating the models themselves has closed.
It’s really important to distinguish models from fine-tunes, which is a key point that was entirely omitted in the Congressional testimony. Regulating models themselves won’t change the impact that fine tuning has in terms of potential harm AI can do.
Let me explain in terms of pizza. Building a model is like baking a pizza from scratch – and I mean, all the way from scratch. You have to grow a field of wheat, harvest it, dry it, mill it, and turn it into flour. You have to grow tomatoes. You have to mine salt. You have to dig a well for water. You have to grow a field of sugar cane or raise bees or something for the sweetener if you use that. You have to grow grazing land to raise cows to get the milk you need for cheese. Baking a pizza from literal scratch would be an enormous, expensive enterprise.
A fine-tune is like taking the pizza that already exists, like one of those store-bought pizza kits, and adjusting the proportions of the ingredients. Maybe you add a bunch of chili flakes to it – that will dramatically change the pizza and how it tastes, but it doesn’t change the underlying model very much. You can do a lot of harm to someone by giving them a super spicy pizza, even if the base pizza was harmless, or giving them a pizza with toppings that they’re allergic to. The base pizza isn’t to blame, but it was part of the delivery mechanism of the harmful impact.
Here’s why this is important. Building a model is incredibly resource intensive. You need massive amounts of compute power, time, properly-labeled data, and human resources to produce the base model. This limits the production of these large language models to big tech companies. On the other hand, fine-tuning a model can be done by you or me with nothing more than a gaming laptop. Going back to our pizza analogy, it’s the difference between the entire supply chain needed to make the pizza, and you or me just sprinkling a handful of store-bought chili flakes on it.
The potential for harm can come from the model, or it can come from the fine tuning of the model. Regulating models will in no way solve the fine tuning issue, and there isn’t a legitimately sensible way to do so that doesn’t basically require government surveillance of literally everyone.
Why? Because some of the best models now are open-source models, models that literally anyone – you, me, the dog – can download. You can download them for free and use them today, and they’re very good as is, but you can also fine tune them on your own to do exactly what you want them to do. In terms of regulating models, the horse has left the barn.
So that key takeaway – that the powers that be are discussing regulating something that’s already happened and can’t be taken back – is critical to understanding where the government – in this case, the USA government – is in their understanding of AI. The USA is behind, far behind the EU, and far behind the tech community, and they need to catch up quickly or else they’ll be legislating for problems that no longer exist.
The second major area where there was a lot of discussion was around liability. We’ve established now that AI created content is, in the USA, ineligible for copyright because it was not made by humans, and copyright law applies only to human-made creations. The big question now is, who is liable for an AI model’s output? We have a couple of precedents here that we could look to, and none of them are an exact fit.
Full disclosure, I am not and have never been a lawyer, and I cannot give legal advice. If you need a lawyer who specializes in AI, go look up my friend Ruth Carter. They do AI law.
The first precedent is the copyright one. Because machines are ineligible for copyright, this implies that their output has no rights, and in a sense then no responsibility for what they create either. This makes a good deal of sense. If a machine spits out, say, racist content, by itself it hasn’t done anything wrong. Someone else today has to take that content and publish it, distribute it, do something with it, and it’s that action which could be in violation of the law.
The second precedent, and one which came up a lot in the hearings, was Section 230 of the Communications Decency Act. This law essentially indemnifies carriers for the content that goes over their networks. For example, T-Mobile, my mobile company, has no legal responsibility for what I do with my devices on their network. If I distribute illegal content, they cannot be sued for my actions. This act is what has caused social media to be such a dumpster fire; companies like Twitter and Facebook have no legal liability for what people post on those networks. In the USA, the Supreme Court just upheld this, so there’s little chance of that changing any time soon.
So when a machine does something wrong, who owns the mistake? The current thinking – unsurprisingly by big tech companies – is that they are not at fault for what their models create. I can see this point; an automaker is not liable for an accident that I cause unless it can be proven that there’s some defect in the car or the car maker failed to warn vehicle owners that doing something dumb would cause a crash. However, the loophole there is that automakers have safety standards they have to adhere to. AI does not, and thus, the comparison of AI models to automakers isn’t really compelling. If we had standards for which models had to comply, then you could indemnify AI model creators if someone used that model in a way that was not intended.
The law around AI in general is still largely unsettled and will definitely change over time; right now, no one really has solid answers to much of anything. The key takeaway for us as end users of AI is to treat it like a chainsaw. Ask yourself the golden question of AI: what could go wrong? What are the risks if an AI deployment goes off the rails? Just as it’s a bad idea to use a chainsaw to, say, do surgery, there are plenty of use cases where you shouldn’t use AI, like hiring and firing.
Speaking of which, employment was another major area where the folks asking the questions didn’t really know what the questions were that they were asking, and even the AI experts didn’t have solid answers. No one does, though economists estimate between 30-50% of jobs will be impacted, perhaps even lost to AI over time, as well as creation of lots of new jobs, most of which we can’t even imagine right now. I’m a lot more optimistic about this right now than I was a few weeks ago.
Here’s why: the invention of the cotton gin by Eli Whitney in 1793 made cotton go from a pain in the ass crop to a hugely profitable one. The net effect of the cotton gin was a dramatic increase in the farming and picking of cotton, powered mostly through slavery in the Southern USA. That’s right – a technological change created a massive boom in the slave trade (which to be clear is awful).
But the key point is that an asymmetry in labor in part of the supply chain had dramatic effects on the rest of it (as well as terrible human costs). It’s probable that we’ll see AI impacts having asymmetric labor effects as well. Think about it for a second; if we mandate, even internally, that human editors need to fact check what AI is creating, then yes, we lose a lot of writers. But as AI scales up, we suddenly need a lot more editors. These are ordered effects; the first order effect is to reduce the number of writers. The second order effects in this example is to increase the number of editors because instead of having 10 articles a day to edit, editors suddenly have 10,000.
This is a critical point to think about in your own information supply chain: if you use AI to scale certain parts, where are the next logical bottlenecks that you’ll need more resources to successfully harness the outputs of AI?
The final area of discussion, and one that was largely fruitless, was about AI safely and morals. This is an area fraught with problems because no one can agree on what is moral. Think about it for a second. Even in a relatively homogenous culture, there are still major disagreements about what is right and wrong. Whose morals are correct? Christians? Muslims? Buddhists? Atheists? Satanists? Who decides what is right and wrong? We live in a world now where there’s such hyper-partisanship and polarization of opinion on literally everything that we can’t agree on anything. We fight over cartoon depictions of candy, for goodness’ sake.
What we do know about AI models is that they’re trained on our data. Copyrighted or not, if it’s publicly visible, at least one of the major AI models has been trained on it. That means that all our foibles and flaws are in these models as well. Everything good about us, everything bad about us, everything that encompasses humanity is in these models to some degree – and that means vastly conflicting morals. It’s impossible and will remain impossible for us to create these same universal AI models that have any kind of morality – especially as we continue to churn out more and more disinformation.
For example, Russian propagandists are doing their level best to pollute the Internet with as much anti-Ukrainian content as possible. Hundreds of attempts by Russian saboteurs have been made to create code in Twitter’s now open-source recommendation algorithms to classify anything Ukrainian as government-sponsored propaganda and reduce its visibility. Some of that garbage – and it is garbage, let’s be clear – will inevitably find its way into large language models, the same way that other hate speech does.
What’s the solution here? This is one area where the witnesses and the elected officials were in general agreement, and I’m in agreement with them: radical transparency. If an organization is publishing an AI model, it must disclose fully and publicly what that model was trained on in a very granular fashion. Not “trained on thousands of books”, but the specific books and editions. Not “social media discussions”, but which specific posts.
We don’t accept nutrition labels any more, especially in places like the EU, where you just don’t bother disclosing information. You’re required to disclose specifics. The same should be true of AI models as well as fine-tuned models. Someone who’s doing fine-tuning should equally be required, if the model is going to be made available for commercial or public use, to disclose everything in the fine tuning dataset so that we can all see exactly what the model is learning.
This is how we’ll solve some of the liability issues around AI as well. Right now, we don’t know how models were trained, so we can’t realistically say whether a model should be liable for its output. But if we require full disclosure of the data a model was trained on, we can absolutely hold accountable a tech company for training on content that’s harmful, false, etc. We could mandate, for example, the exclusion of patently false and wrong information (like content claiming the Earth is flat when it is verifiably not flat) – and companies which do not exclude that information in their training datasets would be more liable for the things their models do wrong.
This is where some of the cottage industries are going to spring up around AI, opportunities for businesses and savvy entrepreneurs to make a lot of money:
- There’s money to be made, especially for folks who have backgrounds in DEI (diversity, equity, and inclusion), to help audit models – especially the training data that goes into models.
- There’s money to be made in the auditing processes themselves.
- There’s money to be made in monitoring models and doing independent third party validation of model outputs.
- There’s HUGE money to be made in curating training datasets that meet specific standards – voluntary standards at first, until the industry or the government gets it together.
- There’s money to be made in the national security and policy implications of widespread use of large language models, particularly around propaganda and disinformation.
AI is an incredibly powerful tool that has no manual and no guidelines right now. If we want to continue making use of its power, we need to better understand its capabilities and regulate the inputs and outputs – what goes into making AI and how people use it – for us to succeed with it in the long term. As we have seen with hostile foreign powers like Russia, there are already attempts to use it to subvert nations and cause tremendous damage with it, so the sooner we figure things out, the better.
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 the livestream bake-off I did this week to test out 4 different generative AI models. I was really surprised at the results!
- So What? Q2 2023 Generative AI Bake-off
- Almost Timely News, May 14, 2023: Resistance to AI
- You Ask, I Answer: Company Claims of AI Generated Content?
- You Ask, I Answer: How to Regulate Generative AI?
- You Ask, I Answer: Large Language Model Capability Limits?
- You Ask, I Answer: Sales Roles Impacted by ChatGPT?
- You Ask, I Answer: Detection of AI Content?
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.
- ⭐️ The Marketing Singularity: How Generative AI Means the End of Marketing As We Knew It
- Powering Up Your LinkedIn Profile (For Job Hunters) 2023 Edition
- Measurement Strategies for Agencies
- Empower Your Marketing With Private Social Media Communities
- Exploratory Data Analysis: The Missing Ingredient for AI
- How AI is Changing Marketing, 2022 Edition
- How to Prove Social Media ROI
- Proving Social Media ROI
- Paradise by the Analytics Dashboard Light: How to Create Impactful Dashboards and Reports
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 recent open positions, and check out the Slack group for the comprehensive list.
- Data Scientist Large Language Model Generative Ai at Pivotal Life Sciences
- Data Scientist at HolidayCheck
- Digital Marketing Analyst at Microcenter
- Machine Learning Engineer – Pytorch Image Modeling at Tower 33
- Research Scientist at Newark Board of Education
- Senior Data Scientist at Flagship
- Senior Digital Analyst at NPR
- Senior Digital Analytics Implementation Engineer at Virgin Media
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.
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
- Developing a consistent brand voice across social media channels via Agility PR Solutions
- Montana becomes first US state to issue full TikTok ban
- Experts Share How to Find the Sweet Spot for Social Media Marketing
Media and Content
- How to Create Conversion Content (That Actually Converts)
- 9 steps to developing a successful lead management strategy via Agility PR Solutions
- How Gen Z is rewriting the beauty industry rulebook and what brands will have to do to reinvent themselves in order to remain relevant via Agility PR Solutions
SEO, Google, and Paid Media
- Single-Page Websites and SEO: The Essential Guide
- 9 DIY SEO Software Tools & How to Use Them Like A Pro
- Google’s Helpful Content Update Can Demote & Promote Content
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!
Tools, Machine Learning, and AI
- Breaking Down AutoGPT via KDnuggets
- Can We Trust AI Decision-Making in Cybersecurity? via ReadWrite
- DeepBrain AI uses virtual humans and generative AI to automate job interviews via SiliconANGLE
Analytics, Stats, and Data Science
- How Does ChatGPT Work: From Pretraining to RLHF
- Fashioning the Future of AI with GANs via Analytics Vidhya
- A Beginners Guide to Anomaly Detection Techniques in Data Science via KDnuggets
- Microsoft Invests in No-Code App Builder Builder.ai
All Things IBM
- How process mining improves IT service management to save your business time and money via IBM Blog
- IBM snaps up Israeli cloud data security start-up Polar Security
- At the Senate’s first AI hearing, lawmakers and OpenAI, IBM execs weigh risk and regulation via Digiday
Dealer’s Choice : Random Stuff
- Tech Stocks Could Tumble As Earnings Recession Sets in: Morgan Stanley
- Debt Ceiling Drama and Rate Hikes Could Send Stocks Toward Sell-Off
- Downbeat Economy Has Left Truck Drivers Facing a Slowdown Like 2008
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.
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:
- My blog – daily videos, blog posts, and podcast episodes
- My YouTube channel – daily videos, conference talks, and all things video
- My company, Trust Insights – marketing analytics help
- My podcast, Marketing over Coffee – weekly episodes of what’s worth noting in marketing
- My second podcast, In-Ear Insights – the Trust Insights weekly podcast focused on data and analytics
- On Twitter – multiple daily updates of marketing news
- On LinkedIn – daily videos and news
- On Instagram – personal photos and travels
- My free Slack discussion forum, Analytics for Marketers – open conversations about marketing and analytics
Events I’ll Be At
Here’s where I’m speaking and attending. Say hi if you’re at an event also:
- B2B Ignite, Chicago, May 2023
- MAICON, Cleveland, July 2023
- ISBM, Chicago, September 2023
- Content Marketing World, DC, 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.
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.
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:
- How to Set Your Public Speaking Fee
- Understand the Meaning of Metrics
- Marketing Data Science: Introduction to Data Blending
- How To Set Your Consulting Billing Rates and Fees
- Is Social Listening Useful?
Want to read more like this from Christopher Penn? Get updates here:
Get your copy of AI For Marketers