PRODUCTHEAD: AI is word of the year
PRODUCTHEAD is a regular newsletter of product management goodness,
curated by Jock Busuttil.
ok product #
every PRODUCTHEAD edition is online for you to refer back to
Anyone can access generative AI, so simply using it is not a competitive advantage
SEO has long sought to game search engine rankings; AI provides a new tool for doing so
LLMs trained on AI output become increasingly detached from reality: “model collapse”
When generative AI is used to create content primarily to game another algorithm that ranks it, the closed, adversarial loop renders the system useless to people also using it. We’re already seeing this with Google search; we’re just about to see it with job applications on LinkedIn. The end-result? Real people will go elsewhere to find what they’re looking for.
I can’t believe it’s only been 10 months #
Collins Dictionary has named ‘AI’ word of the year for 2023. They’ve called it early — I feel they’ve done a disservice to all those words yet to make their mark on the world in November and December. But I digress.
It seems longer ago, but it was only in January 2023 that generative AI exploded into the public consciousness. I was pleased; finally NFTs and cryptocurrency were finally old news.
There must have been a bunch of miffed experts in machine learning, natural language processing and other related fields. They’d been plugging away for years, only to have their thunder stolen spectacularly by OpenAI and others. On the plus side, with all the hype it should at least be a bit easier for them to secure investment in the future.
Fast forward ten months, and what wonders has AI wrought? Some good stuff, some bad, and rather a lot of mediocrity.
Google’s core business is under attack #
Generative AI has a problem with words. They lack soul. AI writing is depressingly well-suited to content farms, business communication and job applications.
Humans are also perfectly capable of bland writing; that’s arguably how we got here in the first place. Even before AI, when you searched for pretty much any topic on Google, the prized top ten non-sponsored results would be dominated by middle-of-the-road articles on corporate blogs, written by unnamed individuals.
Content farms have long existed to churn out articles on given topics primarily as SEO fodder. The writers were typically paid a a criminally small fee, and the quality of content mattered far less than the quantity of words and the appearance of being written by a human being. The content exists primarily to game Google’s search algorithm, in the hope of receiving valuable traffic and consumer engagement in return.
It is this mediocre-middle, SEO-focused content that has likely been a significant proportion of the source material on which ChatGPT and other large language models (LLMs) have been trained. It’s no surprise that generative AIs write so blandly.
However it’s the incentive feedback loop (write more plausible content, rank higher in Google search results) that becomes particularly pernicious when algorithm meets algorithm. Replace the human writers in content farms with ChatGPT and its ilk, and everything can operate at greater scale, 24 hours a day, with no productivity-sapping breaks to eat, drink or sleep. Hooray.
This merely accelerates the race to the bottom. When anyone — not just companies with a healthy SEO budget — can churn out plausible content automatically, it becomes far more difficult for any one publisher to dominate the search results. More article choices, all equally mediocre. It also destroys the value to users of Google search, with the good content drowned in the overwhelming sea of mediocrity.
Google will inevitably tweak their ranking algorithms to try and improve the situation. They have an advertising business to protect. Content farms will respond in real-time by tweaking their generative AIs. Search will remain broken, so real people seeking useful and relevant information will start to find it in other ways, and Google’s search advertising revenue will start to collapse.
Microsoft is about to break LinkedIn in the same way #
We’re going to see the same race to the bottom with job applications. Back in February, I jokingly predicted how AI in various Microsoft products, including LinkedIn, would tender your resignation and apply for your next job on your behalf.
We’re basically living that reality now. LinkedIn already provides AI assistance to write job seekers’ profiles, job descriptions and recruitment (wonderful, more recruiter spam). And as of 1 November 2023, LinkedIn has started offering premium subscribers an AI ‘coach’ to automate how they apply for a job.
To begin with, this creates a two-tier system: applicants assisted by AI arguably have an advantage over those who are not. For those using AI, the effort to apply for jobs is minimised. More automation will mean vastly more candidates applying per job.
To cope with the increase in volume, hiring companies will inevitably resort to their own AI to screen CVs (résumés) and profile suitable candidates. AI-generated content being parsed by another algorithm — sounds rather like the SEO content farms above, doesn’t it?
The quantity of candidates per advertised job will rise, and the overall quality will dip. There will still be the good candidates in there somewhere, but they’ll be drowned out by the increased numbers of the AI-assisted mediocre middle. And with the applicant and recruiter systems both automated in an adversarial feedback loop, changes on the recruiter side to try to favour more distinctive characteristics will only be met by rapid adaptation by the job application generators, which will continue to game the selection algorithm.
Perhaps the mediocre middle is what companies want. Why take the risk of making a bad hire by taking a chance on a candidate that stands apart from the rest? An increase in the supply of candidates only tilts the market further in the direction of the hiring companies, increasing their negotiating power. I would not be surprised if lower salaries are offered as a result.
Meanwhile, expect more enterprising candidates to use their network of contacts to obtain a referral or recommendation from someone known to the hiring manager, in the hope of bypassing the AI-driven CV sift entirely. And as a friend astutely observed, this would only bias the hiring system towards only those with well-established networks of contacts. (It’s who you know, not what you know.)
Final thoughts #
Whether searching for information or job candidates, the next billion dollar problem to solve will be how to cut more effectively through the mediocrity to the good stuff.
Rather than evolving their business, I would expect many of the big players to attempt to preserve their existing, outmoded business models by fighting AI with AI.
When the real innovation comes to offer an alternative, I bet it won’t come from Google or Microsoft, but from a crazy-sounding startup with big ideas.
Speak to you soon,
what to think about this week
LLMs remove a key competitive advantage of publishers. You need to find a new one.
As the public begins to believe Google isn’t as useful anymore, what happens to the cottage industry of search engine optimization experts who struck content oil and smeared it all over the web? Well, they find a new way to get rich and keep the party going.
[Amanda Chicago Lewis / The Verge]
New research suggests that we might have reached a tipping point: AI learning from AI-generated content. This AI ouroboros—a serpent eating its own tail—could end quite badly.
[Jose Antonio Lanz / Decrypt Emerge]
Job adverts present a chicken-and-egg problem: they all need you to have product management experience to secure a job, but you don’t yet have a product management job to gain that experience.
Don’t let this discourage you!
[I Manage Products]
Recently I was explaining to a client why I focus my efforts on finding “force multipliers”. These are what I call activities that allow us to extract multiple benefits from a single piece of work. You could think of it a little like a workplace fusion reaction, where the output ends up far greater than the input effort.
[I Manage Products]
When the vision and strategy are focused and clear, they allow product managers to prioritise and filter the possible options for their products more easily.
[I Manage Products]
can we help you?
Product People is a product management services company. We can help you through consultancy, training and coaching. Just contact us if you need our help!
Helping people build better products, more successfully, since 2012.
PRODUCTHEAD is a newsletter for product people of all varieties, and is lovingly crafted from an absence of hearing.
Read more from Jock
The Practitioner's Guide To Product Management
by Jock Busuttil
“This is a great book for Product Managers or those considering a career in Product Management.”— Lyndsay Denton