Kathy Gibson is at SingularityU Exponential Finance Summit in Cape Town – Artificial intelligence is outperforming just about every other business – but sometimes it can be difficult to understand how it can transform our own organisation.
Manu Sharma, co-founder and CEO of Labelbox, explains how being able to capture unique insights can help companies in all sectors.
A previous business venture, capturing images of earth from a satellite network captured images of traffic around retail stores and even images of oil reserves to be analysed. This data can even be used to predict the global production of agricultural commodities.
“I realised the world is changing at an incredibly fast rate and there is a big technological shift underway,” Sharma says.
Traditional computing and software – Software 1.0 as Sharma calls it – has brought us to where we are now. But he thinks Software 2.0 is going to be all about AI and deep neural networks.
“Now, we are changing from thinking about problems and solutions, to identifying the problem and feeding the neural network with data; and the network will come up with a solution.”
In Software 2.0, the architecture relies on the neural network making a decision that is then reviewed by humans, and then allowed to continue.
“This means you can solve problems then scale it around the world, limited only by compute and data. In the past, you were limited by the humans you had available.”
Training data is key to this process, Sharma adds. “The key is to collect as much data as possible and expose it to the networks. This means labelled training data is the new gold for organisaitons.”
Organisations need to identify and label their primary data to avoid missing out on these innovations.
Meanwhile, new tools are emerging to teach AI – so no more coding is required.
“In Software 1.0, it could take decades of research to allow computers to reach answers that take very little time for neural networks,” says Sharma.
These neural networks are increasing their efficiency at an amazing rate, he adds. For instance, just five years ago, facial recognition didn’t work well: today it is fast and accurate.
“All of these neural networks are not only getting better, they are also getting cheaper. And they are getting bigger. All still essentially limited only by compute and data.”
AI-enabled organisations are those that adopt AI models that learn from data inside the organization, as well as their human expertise involved in teaching them.
“Every organisation has plenty of data. You have to think about how you are going to use that data, convert it into labelled data sets, and use it to solve problems in your organisation.”
Sharma urges organisations to make AI their competitive advantage. “AI is very transformative, it requires a lot of trial and error, and a lot of focus from your team. I advise you to define a very narrow problem that has high ROI in your business and focus on solving that.
“Secondly, collect the relevant data that can be used to solve the problem,” he says. “Then think about how you will apply human insights to turn it into training data.”
Driving the technology revolution is a massive increase in compute power, together with accessibility democratising data and technology. AI and its use in business is the final piece of the puzzle, says Ashley Anthony, founder and CEO of Isazi Consulting, and a faculty member at SingularityU (SU) focusing on artificial intelligence (AI) and machine learning (ML).
To leverage AI, companies need to transition to being data-driven businesses, he says.
To do this, they must look at simplification to improve cost, speed and accuracy. After this, they should look at expansion – using data to shift purpose, technology and revenue.
Anthony outlines the different types of AI that companies have employed over the years.
At the bottom of the stack is rule-based AI, followed by linear AI.
Intuitive AI is the next iteration, and is a lot more useful in overcoming practical business problems.
However, to ensure better accuracy, augmented AI is needed: the use of machine learning and people to come up with the most accurate solution.
Once data has been accurately captured, it can then be simplified. The next step is expansion, Anthony points out.