Artificial intelligence promise v reality
WE have been and will continue to be surrounded, bombarded with promises of what artificial intelligence (AI) can deliver to our businesses, to government, to society. There is so much hype around the subject. I want to bring reality to the discussion. My goal is to help us understand what the promise is and what is reality.
Indeed, artificial intelligence is disrupting industries, its changing everything fundamentally reconfiguring industries, professions, and lives. AI is enjoying high-profile success stories from self-drive vehicles to consumer electronics.
Early adopters are demonstrating high-impact business outcomes in fraud detection, manufacturing performance optimisation, preventative maintenance, and recommendation engines. Without question they are all built on a solid foundation of data.
This column is about down on the ground business and decision support aspect of AI. First of all, there is no AI without data. In particular without good, relevant data.
The problem is relative to the size of the organisation as there always seems to be a huge unmanageable amount of data whether its in finance, health, retail, telecommunications, education, tourism, law, and many other industries.
Decisionmakers can spend upwards of 80 per cent of the time sorting through the data, looking for outliers, examining the quality of the data, and where the defects are. They are not spending time understanding what's signal and what's noise, there is no time to understand what data is important for the decisions they want to make. Instead, they are spending their time wrestling with data. Not making decisions critical to the organisations goals.
The only way of ensuring you can understand what data is needed for the decisions you want to make is by first extracting data from all the relevant systems of record such as point of sale data, product data, customer account data, inventory systems, client history and then integrating that with data from your systems of engagement such as your Facebook page, your website, and your call centre systems.
And that may seem daunting. But the idea is to design a roadmap to build out your decisionmaking capability incrementally over time, delivering value as you progress.
There is no overnight solution and hence it offers the opportunity of first mover-advantage. AI is no overnight solution. It is a narrow domain of an organisations decision making or decision support environment.
Another important consideration is to not focus on the technology. Some of you who have been in IT development will know the key is to first understand what it is that you want to achieve.
In decision-making AI and analytics you'd also better know what you want to achieve.
Importantly understand what the right business question is, instead of simply delivering answers that do not relate to your business goals.
Otherwise, all you are doing is running reports of what happened yesterday thus limiting your analytical and predictive capability, recreating paper files in electronic format.
The AI promise is the introduction of algorithmic applications that will help identify the signal from the noise.
Some expect AI to automagically give you the answers to help run your business better. That is very promising. But what is the reality?
AI is a marketing term. It means everything and nothing at the same time. AI is this big umbrella, and you can put lots of things, lots of technologies and lots of data driven decision-making tools under this big umbrella.
Just as everybody's products were prefixed with the letter "e" in the past, you may recall the E-Business hype, there is an increasing amount of AI in their product descriptions.
I'm going to discuss Artificial Intelligence technologies in the narrow domain of decision-making or back in the day what was called executive decision support — to enhance decision making to enable prediction and classification.
When I started out in the executive decision support industry in the early 90s it used to be called data mining. Data mining is not cool any longer. Data mining has used the same mathematics and algorithms for decades now being heavily used in AI.
That does not mean AI has no new value, that there is nothing interesting there. No. What it means is that it takes the same mathematics and learnings and automates it.
The insight is not that AI machine learning changes the mathematics, but it is that data mining is now automated.
In the old days, the data-miner, the word we used before data scientist became popular would model and predict things like which customers are going to churn in the next 90 days, which customers will close which accounts by when and why.
And the moment the model is built and put to use; the predictive performance of the model starts to go down over time. Because business doesn't stand still, the data keeps changing.
The data miner or as they are called nowadays the data scientist then tweaks the model and lifts the performance of the model. The cycle starts again.
With AI machine learning we automate that process to keep that predictive performance at a constantly high level without needing the manual intervention of a data scientist. That is the innovation of AI machine learning.
AI is overhyped. Most peoples understanding historically comes from Hollywood and I'm not sure that Hollywood is a particularly accurate source. That's probably not a good thing.
Recently a US politician said that about 35 per cent of the population didn't understand AI, and 65 per cent had some understanding.
I'm hard pressed to believe the 65 per cent. The worry is the 65 per cent who believe they understand but their knowledge of AI is heavily influenced by Hollywood.
And that's not the biggest worry, we can live with that. The problem is with systematic bias and how decisions and recommendations are made.
IDC analysis suggests there will be a huge backlash against AI in the business decision making domain because people do not understand it, do not trust it and so don't want to go near AI.
That's closed-minded thinking because AI has a lot of value to offer. But it will not be accepted in society if its perceived to be biased. Bias is not eliminated because we use mathematics and algorithms.
Humans are still responsible for developing these algorithms and providing the data sets that train the algorithms. This is the biggest issue that is limiting the widespread adoption of AI.
One of the most important things to overcome this is transparency. Too often it is seen as a black box. We know there is some data that goes in and there's some recommendation that comes out. But how did that happen is a mystery.
In 2018 the General Data Protection Regulation (GDPR) a regulation in EU on data protection and privacy in the European Union made it law that there has to be explainability, more transparency because once there's transparency then you can examine, question and refine your understanding. But not when people feel there is something hidden.
As a result we're seeing less of a concentration on making AI models more and more accurate but an increasing concentration on making it more and more explainable.
MIT and Stanford are focusing on explainability. How do I make it more explainable to the layperson?
How can they better understand how a decision was made even if they do not agree with the decision, but the recommendation is more explainable.
There’s a perception perpetuated in Silicon Valley that If there’s a new technology the Europeans will regulate it and the Americans will monetise it.
Regulation is not a bad thing, but innovation requires a culture where we are willing to forgive and learn from failure. Failure needs to be part of the learning process and not punish people for failure as an opportunity to learn.
Naleen Nageshwar is a practitioner of executive decision support, data analytics and digital business transformation specialising in business imperatives that can be supported through analytic insights and big data. He runs his own consultancy Data4Digital. He can be reached at naleen@data4digital.com or via his website www.data4digital.com