AI: Expectations vs reality in 2022

Insights and recommendations based upon learnings from the past decade in AI productization

Expectations

Humanity’s imagination regarding AI has always been thrilled to race towards the possible and show it to the world.

Humanity’s recent experience with surging technology innovations is that they happen fast and it’s a race to be first.

Between these two points, the expectations for AI were unspoken, but set:

• It will be unlike anything we’ve ever seen before, literally changing our lives and evolving humankind
• It will happen as fast as past technologies, creating a new booming tech economy

Knowing what we know now from the last 10 years of pursuing the delivery of best-in-class AI experiences, here’s the truth about AI delivery:

The timelines for technology innovations and investment returns we have seen in the past 30 years do not reflect the complexities of delivering best-in-class AI. Although we are able to demonstrate machine learning and natural language processing with successes like Watson Jeopardy and AlphaGo, those were petri dishes of customized data in a controlled setting. To scale what people saw during those broadcasts to the world will require a long-term investment in trained data, subject matter experts, and AI technologists, developers and designers. It will take at least a decade for the world to prepare their data and train an AI supporting workforce.

Reality

In short, it’s a cascading series of chicken-and-egg dependencies.

Roadmaps are prioritized based on predictable business impact.

There’s little to no reportable, repeatable proof of AI’s business impact because there’s no standardized process for gathering impact metrics specific to AI features. Therefore, getting AI prioritized on roadmaps is asking product leaders to take a leap of faith that it will deliver revenue…someday. Maybe.

Customers won’t buy what they can’t see.

In order to gather the impact numbers necessary to get AI prioritized on roadmaps, customers need at the very least to see working AI features that they can try before they buy. This requires AI models to be trained with data relevant to the customer which is often legally prohibited or doesn’t exist yet. Difficult, but not insurmountable with creative thinking around seller demos.

AI timelines cannot be the same as other technologies.

The technologies that created previous booms and established our expectations — websites, social media, online shopping, digital audio and video tools, online collaboration, online community, etc — were are all less complex (compared to AI) and therefore, easier to commoditize. Once we had computers, the internet and smart phones, the next step was to imagine it, build it, watch it go — a widget (remember those???), a website, an app, a service.

AI requires a much larger, longer term investment into research and resources than those technologies.

Why? Because at its core, AI is…

• a chaining together of many technologies that have taken us decades to develop,
• has counter dependencies on computer processing speed and the maturation of machine learning algorithms,
• must be fueled by enormous amounts of human-trained, re-trained and re-re-trained data to provide accurate results,
• and is reliant upon being able to interpret the sum of direct and indirect human and system feedback in order to deliver explainable, trustworthy predictions, insights and automations faster than humanly possible.

This is why AI winters are a thing. It takes a breakthrough in technology, then a catching up of trained data, then finding a repeatable deploy + improve process, and renewed/expanded team skills to support the AI lifecycle. Over and over. Existing AI predictions and recommendations aren’t popping up in every app and tool you use because 50 years ago, no company (or career-seeking college applicant for that matter) was thinking…

What we really need to do is get all those boxes of printed ledgers and surveys and reports that have been sitting in the basement digitized and analyzed so we can use computers to find patterns and respond with human-like predictions and recommendations.

Surviving the winter

The AI vision and progress that has been made will continue for decades to come. We are emerging from an “AI spring” with Watson Jeopardy and other breakthroughs like AlphaGo, IBM’s Project Debater, Apple’s Siri, Google Translate, self-driving cars, automated call centers…the list goes on and on.

This 10 year surge has landed us squarely in another AI winter — or possibly fall — as companies and policy makers come to understand that in order to deliver responsible, valuable AI, they need to hire new people on their product development and legal teams, get their data in order, and wait as computer processing accelerates or quantum computing goes mainstream.

For companies who hit pause during this time, waiting until AI becomes faster and cheaper to deliver, it will be too late to catch up with those who have invested in the long game. Their organizations will be structured to support AI processes and people, while everyone else will be forced to buy from them.

We can take lessons we’ve learned from past tech revolutions to better plan and streamline AI’s trajectory, such as…

• Prototype more, develop less, fail faster
• Connect with users, ask the right questions, get quantifiable metrics
• Put researchers, strategists, designers, developers and especially data scientists side-by-side to build a vision from end-to-end
• Deliver MVP’s, not perfection

But most of all, it’s time to be honest with our leaders, our teams and our investors about the WHY, WHAT, HOW and WHEN behind every AI initiative.

Specifically:

1. Design + research: We need to put AI concepts in front of customers instead of finished solutions so we can make data-informed decisions on where money and resources should be focused to impact both immediate and broad AI progress.
2. Measured impact: We need user researchers, product managers, marketing, and sales to be aligned in gathering and analyzing impact of deployed and marketed AI features, as well as competitor deployments, and deliver insights back to research, product teams and decision makers.
3. Scale: We need to invest in growing the skills, standards and processes of our teams to reflect sustainable, trustworthy, measurable AI lifecycles.
4. Mindshare: We need sales and marketing to have a stake in bringing awareness to deployed AI features and capturing AI specific impact to revenue and marketshare.
5. Realistic strategies: We need to reset expectations on investment and return timelines, identify revenue targets sitting further out on the horizon, and recognize that this is a long game — and it’s not going away.

Every department of an AI investing organization has a critical role to play, and every member of those organizations has an opportunity to innovate and contribute if we want AI to continue creating a new future for the world, as it has for the past 80 years.

Jennifer Aue is a Design Director and Distinguished Designer for AI Transformation at IBM, based in Austin, TX. The above article is personal and does not necessarily represent IBM’s positions, strategies or opinions.