Five Ways to Help Enterprises Succeed with Artificial Intelligence
By Dave Timm, Head of AI Projects
Recently, I attended the AI World Conference in Boston, USA. Attracting world-leading industry speakers and more than 2,000 delegates, the conference focused on how to use AI to drive business value.
The conference gave good insight into where corporates are currently at with AI, as well as practical tips for increasing your chance of success when first starting out. Below, are my five key takeaways.
1. For enterprises starting out with AI, focus on Supervised Learning first
Machine learning, in very simple terms, is the ability of computer systems to “learn” from data without being explicitly programmed, and is already being used by enterprises for a range of applications, such as personalising customer service, targeted advertising, or calculating risk.
Amongst enterprises we’re seeing lots of examples of Supervised Learning (using a known answer to train the model), some good examples of Reinforcement Learning (where a user’s feedback is used to train the model), yet hardly any real-life examples of Unsupervised Learning (“here is a set of data, please make sense of it”).
For companies looking to implement machine learning AI, it makes sense to focus on Supervised Learning first. I recommend enterprises find a set of structured, labelled data in an area of business value and develop a Supervised Learning model. Don’t lose sight of the importance of domain expertise, and ensure you have access to people who are not only aware of machine learning technologies, but also know how to apply them to solve a business problem.
The next step is to look at including machine learning models that perform a specific task. We are seeing a proliferation of models, which are also becoming increasingly specialised. The trick therefore is to manage and orchestrate multiple services to do a specific job. For example, a photograph or video may be analysed by several focussed models to get specific pieces of information, such as recognising what brand of shoe or jeans a person is wearing. This is set to have a huge impact on targeted advertising.
2. Understand issues around transparency, bias and trust
There is a push from many enterprises already using machine learning models, in particular deep neural networks, to start focusing on improving the transparency of the models. These models ‘learn’ from their data and don’t always provide auditable answers. As a result, risk and audit departments are leading the trend in favour of more transparent models. Companies looking to introduce AI will need to make decisions about their own risk appetite. Whilst deep learning neural nets can give higher value, they are less transparent than other models.
3. Consider how conversational interfaces, and particularly voice, can work for you
There’s an overarching excitement in the industry for conversational interfaces. Here, users communicate with the software via natural language. Natural Language Processing is used to understand the intent, and then take action. What started out as ‘chatbots’ is now becoming AI-enabled, personalised user interfaces that interact with multiple other systems.
The majority of companies are achieving conversational interface with text. Whilst voice is emerging in the home and enterprises, it requires custom dictionaries, bespoke language models and sometimes acoustic modelling too. Although complex, voice recognition is improving rapidly and we expect this area will be huge in the coming year.
With a conversational interface an end user can simply talk and the software has the intelligence to fulfil the requests. It requires minimal training, no need to navigate multiple systems, and leads to a very different and much simplified user experience.
4. Improve your chances of business success with these strategies
There were many stories of success shared at the conference, as well as an acknowledgement that AI projects are difficult and failure is a reality.
Here are some key takeaways for a successful AI project.
- Never underestimate the value of good change management, especially up-skilling the executive level on what is possible.
- Spend time defining what success will look like. It reduces the risk of the expectations and delivery not matching. A large amount of AI projects start as proof-of-concept and don’t have the upfront discipline of success criteria.
- Success depends on solving a specific business problem from a customer centric perspective.
- To achieve AI success you need domain expertise, not just tech skills.
- One presenter suggested that you have to see tangible steps towards ROI within 6 months. If not by then, it’s probably not going to happen.
5. AI will amplify human capability, but be aware that there will be significant jobs impact
While there’s a huge amount of optimism for the future of AI, and a message that AI will augment and amplify human capability, it will also clearly impact employment.
One analyst firm, Forrester, presented a consensus view that there will be a net 7% loss of jobs within 10 years. More specifically this is broken down as a 17% job removal and 10% job creation. Whilst a 7% loss doesn’t sound too bad, keep in mind that it is consistent with the jobs impact of the banking crisis in 2008 over a prolonged period.
Therefore, organisations will need to evolve as AI technology does. Companies will need to consider the ways in which humans and robots work together, and how their organisational structure and operating model supports that interaction.
It’s clear that 2018 will be a big year for AI, with many predicting that the technology will live up to the hype. Investment will continue, and big players will continue to make acquisitions, often to gain more data, and models will become increasingly more available.
There are now plenty of examples of success in enterprises, but also plenty of lessons to learn to improve chances of success. Conversational Interfaces will dominate, machine learning will come of age, but transparency will be an issue. And overall, the employment risk is real, and work is needed to continue promoting that risk.