The State of Enterprise AI in 2017

Over 1,400 delegates gathered this week in London to discuss how AI is impacting their business and where they see the technology headed as a part of AI Summit 2017. The AI Summit is organized by AI Business and is the only conference in the world dedicated to investigating the practical implications of AI for enterprise organizations.

Participants in the conference skewed European given it’s London locale, but major players – including Google, Microsoft, IBM, Intel, NVIDIA, and Salesforce – all made an appearance and shared strategies about how to approach AI within the enterprise. I could track a few themes throughout the two days of talks and conversations on the exhibit floor:

  • Speed and Scale: How fast are things moving? How big is AI going to be?
  • Hype vs Reality: Is AI really going to change everything? Where are we in that process?
  • Responsibility: Who builds AI? Who uses it? What happens people whose jobs AI replaces?
  • Deployment: How do you plan a project? How do you track success?

Speed and Scale

There was widespread consensus that developments and implementations in AI are moving very quickly and that companies who hadn’t already started planning their approach to AI need to start now or risk missing the transition.

Some areas will move faster than others. For example, regulated industries like healthcare and insurance will need explainable approaches with provable causation while ad tech and retail companies can implement methods that just rely on correlation.

Hype vs Reality

There are companies making enormous bets that AI is the future and that we are close enough to capitalize on the promise. However, the question “When are we going to see something besides chat bots?” came up in multiple conversations and there was a sense of frustration with the marketing image of AI overpromising relative to the AI you can actually touch and feel today.

That hasn’t prevented companies like Philips going all-in and promising that within five years every single one of their products will be powered by AI according to Chief Innovation & Strategy Officer Jeroen Tas.

It doesn’t help that most AI solutions are built around the same algorithms. Algorithmic innovation starts in academic research and then makes its way into industry. Industry players can leverage the practical tooling available now using their domain expertise, but it’s not clear how much differentiation there will be if they aren’t investing in novel approaches, either through research or acquisition.

Responsibility

The press is fascinated by the ramifications of AI automating jobs out of existence and many people raised these issues as a part of their talks. Opinions ranged on the exact affect AI will have on jobs, with some thinking that there will be as many jobs created by the new AI ecosystem as there will be replaced. However, that won’t address the need for new skills. Will people be able to transition from jobs that rely on rote skills to those that are enabled by AI? What does an AI-enabled human job look like?

Nobody addressed the proposed solutions to this potential problem, things like a guaranteed basic income or a job replacement tax, at least not in conversations that I was a part of. Clearly it’s an area that needs more thought and more discussion.

Many people were also keen to raise the topic of diversity, both in the creation of AI and access to the technology once it has been deployed. It is clear that we need to expand the pool of people who are working on creating AI – it was noted that only 6 of the 67 speakers were women – in terms of gender and other pluralities of identity.

The question of accessibility to AI technology was raised but no clear answers emerged. The issue was brought up early at the very first talk of the conference by Jon Andrews of PricewaterhouseCoopers who spent a good portion of his time covering the human role in AI. Though we don’t yet have answers here, it is vital that we think very carefully about the potential ramifications of the technology we are building.

On that front, we already have some early thought leaders in the space who are helping to clarify the rules. Jeremy Silver at Digital Catapult highlighted groups focusing on the ethical challenges raised by AI. And, of course, great advice from the next generation:  “Bad people shouldn’t build robots”.

Deployment

For as many grand questions that were raised and remain unanswered, there were some very grounded, tactical pieces of advice that point to an emerging set of best practices for deploying AI-enabled projects in the enterprise. For most organizations right now, these projects are at the periphery as people evaluate how to best integrate, but it is clear that making AI central to the business is a process already underway.

The recommended first step for project planning is to enter into these projects thinking about business objectives. You should not think of this process as “deploying AI” into an organization, so practitioners need to understand what AI is actually capable of and how that intersects with their organization’s business needs.

Proof of Value

Identifying those intersections will allow you to define KPIs that can actually be moved by AI technology, allowing your business objectives to drive internal adoption. This also allows organizations to run “proof of value” projects – everyone agreed we are beyond “proof of concept” when it comes to AI – where the KPIs and success criteria are clearly defined.

A common approach for enabling this is to identify an existing process where AI can drive metrics, for example in reducing labor cost, and then run that process using an AI-enabled approach alongside the existing team, either with or without their involvement. This gives you a side-by-side comparison of the approaches and allows you to quickly gather feedback from stakeholders as you iterate. You can also do this using historical data where the outcomes are already known, for example simulating last year’s compliance reporting process using a new AI-based solution.

Lean on Vendors

Using outside vendors that have purpose-built AI products was a common approach to help scale experience levels internally. Large organizations like Deutsche Telekom are using startups when they provide best-of-class solutions, but they don’t create exclusive relationships and have fallback plans allowing them to migrate if needed.

There are also particular design considerations for AI-driven projects. Noel Lyons of Barclays explored this effectively, discussing whether or not it was better to be proactive or reactive with users – the answer is both depending on the task – and relating how much effort his team spent finding a conversational tone for their chat bot that users enjoyed; users responded negatively to tones that were too robot-like but also did not like tones that were overly-colloquial.

The Bottom Line

Organizations are betting massively on the potential of AI, some of them shifting their entire business to take advantage of the perceived power of this technology.

Though we are clearly still in the early days, things have moved very quickly in the space up to this point and enterprises cannot afford to take a “wait and see” approach to understanding how AI will shape their industry and business going forward.


Also published on Medium.