Innovation Monitor: Gartner Hype Cycle Trend #3 — Formative AI
Welcome to this week’s Innovation Monitor. Formative AI is Gartner’s term for AI that, in simple terms, dynamically adapts based on context. Put even simpler, imagine AI that “creates” things. The term can encompass everything from AI that generates novel content (like This Person Does Not Exist) to AI that assists in drug discovery (see Atomwise) to AI that helps develop software (ex. Kite), and anything in between.
We hear the concerns: AI will replace coders. AI will replace designers. AI will replace “name your job function”. Formative AI represents the algorithms and systems that are now not only engaging in process improvements and analyses, but these technologies can start creating things, by “itself.” So this week, we’ve put together a quick snapshot of three important Formative AI categories:
1) generative design,
2) AI-assisted software development
3) generative adversarial networks
An example of this is OpenAI’s GPT3 that can create slide presentations, write programming code, and parse and translate legalese (which we previously highlighted.) These technologies are shaping a future where many human tasks are influenced and shaped by machines to be faster, more efficient, and more intuitive.
Finally, as TikTok is top of mind for many in media and tech today, I wanted to link back to our earlier edition: The TikTok Backstory as it feels as relevant as ever. As always, we wish you and your community safety, calm and solidarity as we support each other through this unprecedented time. Thank you for reading!
Erica Matsumoto Generative Design While we’re already augmented in the sense that our brains are basically filtered through our smartphones, all the tools we’re using are, when you boil it down… passive. Tools do nothing without our explicit direction — from humanity’s first chisel to AutoCAD, the potential is tied to our expertise. We push, we get a result.
Generative tools only need constraints — they help generate designs based on predefined parameters. For example, you would tell a generative AI that creates drone schematics that you want to have four propellers and you want the device to be as aerodynamically efficient as possible.
Here’s where things get really interesting: these systems come up with designs that are essentially alien to humans thanks to the enormous amounts of data they ingest and zero inhibitions or preconceived notions. Just take a look at the drone and plane partition designs in the 2017 TED video below — they look like they were conjured on some arachnid-populated planet.
But the largest near-term potential isn’t robots that can show us how to win at Go with ethereal strategies or write our news features, but human-machine collaboration. An interplay where each party is helping the other create something more intuitive. To add to that: machines telling other machines how to improve themselves, like a connected car communicating with a city traffic system to determine the best route or better traffic light timing.
This Design News slideshow demonstrates generative designs that made it into the real world, though it doesn’t explicitly state that AI was used. Sophisticated generative AI can create a synergistic interaction between engineer and algorithm, more so than hard-coded programs.
That’s because these algorithms can be trained not just to optimize engineering parameters but aesthetic requirements, commercial impact, and production costs. You can’t do that with hard-coded software. This augmentation speeds up design, development, and even marketing.
Here are two examples from the Accenture piece I just linked (also check out nTopology and Additive Flow for generative software examples):
General Motors became one of the first automotive companies to leverage generative design to reduce the weight of its vehicles. In 2018, the company worked with Autodesk engineers to create 150 new design ideas for a seat bracket and chose a final design that proved 40 percent lighter and 20 percent stronger than the original component.
Under Armour leveraged generative design algorithms to create a shoe with an optimal mix of flexibility and stability for all types of athletic training — inspired by tree roots. The algorithm came up with unconventional geometry that was 3D printed into a shoe and tested by more than 80 athletes in a fraction of the time that it would have taken in the past. AI-assisted Software Development (Software Eats Software) “It’s not that programmers are being replaced by robots — rather, AI-powered tools are making project managers, business analysts, software coders, and testers more productive and more effective.” — Deloitte
Over the past year and change, dozens of AI-based software development tools have entered the market. The spur of innovative products is partly driven by a dilemma many firms face: lack of developer talent. In addition, over half of all software projects are late and over budget. Lack of talent and costly software mistakes (at least $319B in 2018) has opened up a necessity for tools that can augment developer workflows. These augmentations come in a variety of forms (for a deeper dive into each, see the Deloitte link by the quote):
- Project requirements: This is the process of tracking what users need from a piece of software — “a major cause of delayed, costly, or failed projects when done poorly.” Some companies have introduced NLP-powered digital assistants that can analyze requirements docs, flag inconsistencies, and offer improvements.
- Coding and bug detection: AI-powered autocomplete tools like Kite help coders use less keystrokes and focus more on the end-product. Facebook uses a bug detection tool called Aroma that “predicts defects and suggests remedies that are thus far proving correct 80 percent of the time.”
- Deployment: There’s even AI to predict deployment failure by looking at historical data from code releases and application logs.
We can get even deeper here. Machine programming systems generate usable code (so code that creates code). We’ve already seen a functional example with GPT-3:
DeepMind created a system that generates more efficient versions of algorithms, while researchers from Intel, MIT, and GIT developed the Machine Inferred Code Similarity system, which can infer the meaning of a piece of code the way an NLP model would a piece of text. Generative Adversarial Networks Where do generative adversarial networks (GAN) — the tech behind the last few years of bizarre psychedelic AI art and hilarious celebrity impersonations — actually fit in a business?
Outside of the entertainment industry, GAN-based software can actually help with, of all things, financial fraud. American Express is using GAN models to create fake financial data (such as credit card transactions) to help train other algorithms to detect fraudulent activity.
Like with the visual output you’ve seen, the generated financial data is meant to look real, but is a bit off. As Fortune put it, “data with obvious anomalies, such as multiple purchases of toilet paper in New York City on one day, followed by a lawnmower purchase in Bakersfield, Calif., the next, would be less effective.”
Other businesses have used GANs to generate transaction data, notably Amazon, which used a model to create ecommerce transaction data. This data, according to researchers, could be used for “product recommendation, targeting deals, and simulation of future events.” Though, unlike their visual counterparts, it’s difficult to tell if generated transaction data is effective: “the technology is so new that there are no ‘commonly accepted techniques’ that the researchers can use to grade the software,” Amex researchers said. This Week in Business History On September 17th, 1985: Steve Jobs submits his resignation letter to Apple Computer’s vice chairman, A. C. “Mike” Markkula after losing a boardroom battle for control of the company with then-CEO John Sculley (on Sept 16th). The letter below is a fascinating tech time capsule:
This morning’ s papers carried suggestions that Apple is considering removing me as Chairman. I don’t know the source of these reports but they are both misleading to the public and unfair to me.
You will recall that at last Thursday’s Board meeting I stated I had decided to start a new venture and I tendered my resignation as Chairman.
The Board declined to accept my resignation and asked me to defer it for a week. I agreed to do so in light of the encouragement the Board offered with regard to the proposed new venture and the indications that Apple would invest in it. On Friday, after I told John Sculley who would be joining me, he confirmed Apple’s willingness to discuss areas of possible collaboration between Apple and my new venture.
Subsequently the Company appears to be adopting a hostile posture toward me and the new venture. Accordingly, I must insist upon the immediate acceptance of my resignation. I would hope that in any public statement it feels it must issue, the company will make it clear that the decision to resign as Chairman was mine. I find myself both saddened and perplexed by the management’s conduct in this matter which seems to me contrary to Apple’s best interest. Those interests remain a matter of deep concern to me, both because of my past association with Apple and the substantial investment I retain in it.
I continue to hope that calmer voices within the Company may yet be heard. Some Company representatives have said they fear I will use proprietary Apple technology in my new venture. There is no basis for any such concern. If that concern is the real source of Apple’s hostility to the venture, I can allay it.
As you know, the company’s recent reorganization left me with no work to do and no access even to regular management reports. I am but 30 and want still to contribute and achieve.
After what we have accomplished together, I would wish our parting to be both amicable and dignified.
Steven P. Jobs