PRODUCTHEAD: There’s hope for us (and AI) yet

planet product #

---

tl;dr

AI is:

accelerating the design of novel proteins, enabling a new wave of vaccines and drugs

predicting extreme weather events, helping to protect residents

listening to the rainforest and sends real-time alerts for chainsaws, trucks, cars and signs of incursion

monitoring the world’s oceans for illegal fishing activity

---

every PRODUCTHEAD edition is online for you to refer back to

hello

We’re all doomed — and it’s the fault of artificial intelligence. At least according to a recent Reuters/Ipsos poll, anyway.

61% of Americans they surveyed believe that AI poses risks to humanity. (Mind you, I’m certain that the survey could have replaced “AI” with almost anything and would have drawn similar conclusions.)

I don’t share the view that the AI apocalypse will look like Terminator 2: Judgement Day. No, based on the torrent of soulless, vapid and inescapable AI-generated content I’ve encountered this week, I’m worried I’m going to be numbed into submission instead. And yet, I still have hope.

Spotify’s new DJ #

Spotify recently rolled out a new AI-powered DJ. It sounds to me like a cross between Ice-T and Huey Morgan of The Fun Lovin’ Criminals. (Actually its voice model is based on Xavier “X” Jernigan, Spotify’s head of cultural partnerships.)

While it initially scores high on the “sounding like a real human” scale, its charisma soon tails off. Over time the AI voice just becomes a bit monotonous — it lacks dynamism. It’s not particularly annoying, but equally I’m not sure it adds anything to my listening experience.

I do admire its brave and foolhardy segues, however. My favourite was when it leapt from mid-90s indie rockers Terrorvision to Sibelius with a bit of incongruous patter in between. Veeery different energies.

Drowning in Temu #

There are Temu adverts everywhere. They’re both inescapable and inexplicable, peddling all sorts of random tat.

The item descriptions sound like they’re written by a generative AI, and they’re read out by a bored-sounding budget AI voice clone of a satnav circa 2006. So a low score on the “sounding like a real human” scale, then.

The sudden ubiquity of Temu adverts brings to mind the creeping horror of someone relentlessly chasing me while they constantly read out product descriptions in a robotic monotone. Terminator x the Shopping Channel. *shudders*

There’s hope for us yet #

It may feel like the pinnacle of humanity’s achievement with AI has been the creation of unsettling pizza chain adverts and imagined photos from the UK’s coronation afterparty. But there’s a glimmer of hope for us after all.

While they may not attract the same kind of sensationalist press coverage, there are many organisations using machine learning and AI for more altruistic purposes.

Protein origami #

The ability to predict and model viable protein structures may not sound particularly glamorous, but is vital if you’re trying to figure out what an existing protein does, or to design an entirely new one for a specific purpose.

In 2020, DeepMind’s AlphaFold deep learning model revolutionised the approach to a problem that had been vexing computational biologists since the late 1960s.

Even with the right building blocks in the right order, a protein only becomes biologically functional when it’s folded into the correct three-dimensional structure. Figuring out these structures would typically take teams years of painstaking experimental research.

AlphaFold demonstrated it could accurately and relatively quickly predict the complex three-dimensional shape of a protein from its sequence of amino acids, its building blocks. While it doesn’t completely solve the protein-folding problem, its reliably accurate predictions represented a transformational advance.

AlphaFold helps us understand existing proteins, but what about designing new ones from scratch?

Designer vaccines #

Vaccines are designed to cause our bodies to learn a new immune response (create antibodies) targeted at a specific virus. Antibodies latch on to specific protein structures (‘spikes’) on the virus in order to neutralise it. The closer the match between the antibodies and the virus structure, the more effective the vaccine.

The Institute for Protein Design, University of Washington seeks to design novel synthetic proteins for specific purposes, one of which is the creation of vaccines.

As before, a successful vaccine protein needs the right ingredients and folded 3D structure to trigger the production of antibodies that match the target virus closely. And that complex protein structure needs to be able to be synthesised for real in a laboratory.

To figure out these complex protein structures, the Baker Lab team at the Institute for Protein Design have adopted similar diffusion AI techniques as DALL-E and Stable Diffusion use to generate realistic images.

But rather than creating pictures of partying royals, their RF Diffusion model generates specific and viable 3D protein structures, and does so far more quickly than traditional research techniques. This use of deep learning AI opens the door to more rapid development of highly effective vaccines and other drugs, while also reducing the risk of unwanted side-effects.

Final thoughts #

If the constant barrage of terrible AI-generated content is the price to pay for advances in immunology that prevent the next global pandemic, then I can probably live with that. But I do hope Temu’s marketing budget runs out soon.

This week I’ve pulled together some more stories of constructive uses of AI that will restore your faith in humanity.

Speak to you soon,

Jock

---

---

what to think about this week

Thanks to DALL-E, the race to make artificial protein drugs is on

Having solved nearly all protein structures known to biology, AI is now turning to a new challenge: designing proteins from scratch.

Far from an academic pursuit, the endeavor is a potential game-changer for drug discovery. Having the ability to draw up protein drugs for any given target inside the body—such as those triggering cancer growth and spread—could launch a new universe of medicines to tackle our worst medical foes.

A love-child of structure prediction and creative generation

[Shelly Fan / Singularity Hub]

👉 Find out more about Baker Lab at the Institute for Protein Design

After a “red alert,” experts use AI to predict extreme weather

As extreme weather events become more frequent and intense due to climate change, predicting and preparing for these disasters becomes increasingly crucial. In response, experts are turning to AI to revolutionize weather forecasting and better predict extreme weather events.

Jua’s system generates global predictions in less than a minute

[Frederik Bussler / Hacker Noon]

👉 Find out more about Jua (of which Leah Tharin is head of product)

---

---

Saving the rainforest through cloud computing, AI, and machine learning to detect threats

For close to a decade now, San Francisco-based nonprofit conservation technology startup, Rainforest Connection has been capturing millions of hours of sounds from rainforests across the globe.

If disturbances are detected, the nonprofit’s systems automatically send alerts to their on-the-ground partners, who intervene to stop illegal logging and poaching — in real-time.

Detecting the sounds of illegal logging and monitoring biodiversity

[Reethu Ravi / Lampoon]

👉 Find out more about Rainforest Connection

How OceanMind uses satellites and AI to protect the ocean

OceanMind is a nonprofit organization that powers marine enforcement and compliance to protect the ocean’s ability to provide for human wellbeing. Using satellites and artificial intelligence, they help authorities enforce more effectively and industry to work more responsibly. And here to explain in more detail how this work is done and the difference that it is making is Nick Wise, the co-founder and CEO of OceanMind.

Analysing fishing vessel actions in real-time to inform enforcement

[Denver Frederick / The Business of Giving]

👉 Find out more about OceanMind

recent posts

What games taught me about customer onboarding

Video games aren’t necessarily everyone’s cup of tea, but some of the most successful games and products share a common attribute: they help the user become more skilled throughout their journey.

Onboarding is really a continual process

[I Manage Products]

Should I take a product manager job in a sales-led company?

Hi Jock,

I’m currently applying for loads of product manager jobs. I’ve received an offer from a sales-led company where the Product team reports in to Sales. Should I take the job?

Demonstrate good practice AND deliver good product

[I Manage Products]

Control your narrative

Years ago, someone once told me that “perception is reality” when it comes to reputation at work. Of all the lessons I’ve learned in my career, this has been by far one of the hardest.

How to regain control

[I Manage Products]

PRODUCTHEAD is a newsletter for product people of all varieties, and is lovingly crafted from recently rediscovered storage space.