Clearing three misconceptions about AI

A forecast by global market intelligence firm IDC in March 2018 states that worldwide spending on Artificial Intelligence (AI) systems will reach US$19.1 billion (S$26.2 billion) in 2018, an increase of 54.2 per cent over the amount spent in 2017.

Not surprisingly, business leaders are eager to know how AI and Machine Learning (ML) can help them.

However, at the root of this lies a set of misconceptions about technology and its applications.

Misconception 1: AI will replace the human.

This is by far the most common misconception I have encountered. It seems everyone cannot imagine that both the AI and human can co-exist. Nothing is further from the truth and I illustrate that with an example.

Think of an HR hiring system that employs the use of AI. Out of the thousands of applicants for a job posting, do you think the AI system is able to screen out all the wheat from the chaff and zero in on that one single candidate to be recommended for hiring? Clearly in today’s context, a more likely scenario is that from the thousands of applications, the AI system is able to narrow down to the applicants who have the best job fit and highest propensity to accept the offer. These applicants are then passed on to a human HR executive for further screening/interview.

Likewise, if your company is launching an AI product, you should not expect it to fully replace a human. The value of AI comes in automating, making sure that humans are able to be placed out of the picture for as long as possible. Rather than screening thousands of applicants, the HR executive only needs to screen out perhaps a dozen or so, saving the organisation time and money.

Misconception 2: AI systems are just about software and algorithms.

AI systems are very different from traditional Information Technology (IT) systems.

For AI Systems to perform well, a good software is just one of the many pre-requisites, alongside access to useful data. A survey of over 307 digital commerce organisations currently using or piloting AI by research and advisory firm Gartner in October found that the top challenge these companies face when deploying AI systems is a lack of quality training data.

If I may make a comparison, software would be the brain and data would be the education. You might be able to purchase the best software in the world. However, for the software to work, you require engineers to put in the time and energy to tutor the software with the correct type of education (data). Just as a brain is only as sharp as the education it is provided with, the effectiveness of the AI system is only as good as the data that is being fed into it.

With AI systems, more often than not, the data fed into them (at least in the initial stages) is loaded with initial biases. It often takes several iterations before these biases are eliminated. If the data comes from only one source, the AI systems trained will only be as good as that singular source. The growth in the power of computer processing today hence enables AI systems to benefit from being trained from multiple data sources.

Misconception 3: AI systems are complicated, expensive and hard to implement.

AI and ML have progressed by leaps and bounds over the last 10 years. The rise of numerous open source AI frameworks (tensorflow, pytorch, caffe) has simplified the algorithmic complexity for engineers. The prevalence of cloud platforms such as IBM Cloud, Amazon Web Services, Google Cloud Platforms and Microsoft Azure has provided engineers with raw computational power at an affordable price. These two factors aided in the democratisation of such AI and ML technologies. The barrier to implementing such solutions has thus been significantly lowered.

As a personal example, last year I taught my 15-year-old nephew how to create his own convolutional neural network (using tensorflow) in less than 20 hours. He thereafter took this knowledge and created a simple AI engine that, if you feed the engine a photo of his friend, the engine will accurately identify the name of his friend in the photo.

How then should business leaders and their staff prepare for AI? The most practical solution to these misconceptions is early and correct training.

In fact, the Gartner report highlighted that the second challenge facing firms is the lack of appropriate in-house skills.

Although it is widely acknowledged that AI might bring about a loss in jobs, this is likely to be transitory and eventually new jobs will be created.

According to Gartner, the greatest value AI brings occurs when the human is augmented by AI.

Professor Andrew Ng, formerly Baidu’s chief scientist, once famously said that “AI is the new electricity”.

But how much value a business can generate from AI is determined by how well they are integrated within the business processes.

A firm might have access to the best AI system in the world. It may even start off with the correct data.

But it is when the firm is able to integrate both the AI system and data pipeline within the business processes, supported by a team of well-trained workers and business leaders, then the true value of AI can be maximised.

ABOUT THE AUTHOR:

Dr Donny Soh is the Director of Data Group at Future-Moves Group, an international management consultancy. He holds a PhD from Imperial College, London.