putilov_denis - stock.adobe.com
Artificial neural network uses in the enterprise continue to broaden as organizations digitally evolve. Businesses increasingly find it hard to ignore the upside of using neural networks, even despite ethical concerns and implementation challenges with AI.
What are neural networks?
Artificial neural networks are a type of data processing based somewhat on the structure and processes of the human brain. They are comprised of an input layer, output layer and a hidden layer, with backpropagation, an algorithm used for supervised machine learning, a foundational piece of the process.
Organizations largely deploy two types of neural networks: convolutional and recurrent. Each neural network has its own set of uses.
Convolutional neural networks use cases
Convolutional neural networks (CNNs) are best suited for solving problems related to spatial data, such as images. Organizations use them for services such as facial recognition software, analysis of medical results (x-rays) and image classification on retail websites for targeted marketing.
E-commerce sites, such as eBay, utilize CNNs to create a more efficient buying and selling platform, improving customer experience.
"The use of deep learning and visual experiences has been a key focus for us," said Nitzan Mekel-Bobrov, chief AI officer at eBay.
For example, eBay uses neural networks to automatically list products, such as trading cards, for sellers based on data collected from previous listings. Sellers can take a photograph of the cards they want to sell, and the CNN will identify key features such as what type of cards they are and when they were issued, then create the listing for the seller.
"The vision is really to be able to get to a point where a seller could take an image of anything and we'd be able to generate the listing for them automatically," he continued.
Thanks to improvements in technology, use cases for CNNs increasingly involve video components.
CNNs, for example, can extract information from videos to track data such as the number of broken streetlights in a city, said Sreekar Krishna, national leader of artificial intelligence and head of data engineering at KPMG.
The ability to answer these questions using video analysis can benefit insurance companies. CNNs are well-equipped to handle damage analysis, crash reconstruction and other forms of spatial analysis, for example.
"You can reconstruct the damages that are happening on either vehicles or even homes … you can actually analyze what's the impact of the damage by analyzing these videos and images," Krishna said.
Recurrent neural network uses
Meanwhile, the strength of recurrent neural networks (RNNs) lies in their ability to analyze temporal, sequential data such as text and time-series data. Common applications of this technology are in speech recognition and forecasting.
Retail companies and other vendors use RNNs to monitor customer habits, then proactively attempt to retain customers when the RNN detects potential red flags.
For example, RNNs can help identify customers who have made several returns in a short period, said Kirk Borne, chief science officer at DataPrime, Inc., an AI-powered matching site for job candidates and jobs.
"They're probably not going to buy from us ever again," Borne said, adding that neural networks can help answer "what can we do to intervene?"
Solving business problems with neural networks
When implementing neural networks, "you always have to choose a subset of the problem to tackle first. And sometimes that can look like one feature … one type of problem that it could solve," Mekel-Bobrov said.
Business leaders see neural network implementation as a gradual, calculated process.
Iteration is a significant part of defining where to implement neural nets and to what degree. For an e-commerce site such as eBay, this applies when using AI to create a product listing based on similar listings previously posted by vendors. At first, the data set (i.e., previous listings) might start small, so the neural network can only create new listings for commonly sold items.
But as an e-commerce site sells more items and the data set grows, the deep learning-driven product listing technology can create listings for a larger variety of products and with greater accuracy.
Retailers and financial organizations also commonly use neural networks for fraud detection.
A neural network can identify behavior outside of the norm, making note when a person who typically purchases gas once a week buys it multiple times one week, for example.
"It's the combination of the time of the day, the product you bought, the quantity you bought, the location, you bought it -- that kind of weird combination of that says, 'now, this is fraudulent'," Borne said.
The challenges of neural nets
Businesses also need to understand the challenges behind creating and using neural networks. Common problems include a lack of transparency in how a model operates, bias in a model's outputs and not having enough computational resources to run models at scale.
Sreekar KrishnaNational leader of artificial intelligence and head of data engineering, KPMG
For companies operating at a large scale, like eBay, "the computational resources needed are quite challenging … the more progress we're making on infrastructure and compute, the faster we're able to actually move in terms of rolling out deep learning in many different use cases," Mekel-Bobrov said.
A real-world example of the ethical problems that arise from relying on deep learning is determining creditworthiness. If an individual applying for a credit card is declined by a lender that uses neural networks as the basis for their software, their credit history and financial trajectory are determined by a machine, not a human.
The black box nature of AI becomes a problem when the algorithm's decisions impact an individual's life, such as whether they are accepted or declined for a line of credit. A full explanation is needed into why an application passed or failed, so people need to understand how a model came to that decision.
That said, credit lenders increasingly implement neural networks to parse through prospective customers' financial history and determine creditworthiness. Companies are seeing significant benefits from implementing, and iterating, neural networks for recruiting and hiring. The risks associated with neural networks often aren't enough to outweigh the potential improvements to business processes.
How business leaders can use neural networks
When considering neural networks, businesses should first define the parameters for where they want neural networks to influence business and determine where they can live, and can't live, with the results of temporary poor technology performance.
"You have to train the workforce that is interfacing with the technology that the technology is not there to make your decisions … Google gave me a bad search result, big deal. My life did not change. Amazon recommended the bad product, big deal. My life did not change," Krishna said.
The number one question IT leaders need to ask their teams is "Have we measured our value statement before we jump into AI?" Krishna said.
Every industry needs to define its value statement or goals to decide whether implementing artificial neural networks is the right decision.
Krishna noted that "asking which step in the process is the one that is causing the highest pain point in the organization, then attacking that problem using AI" is the necessary mindset when implementing neural networks.
These inherent risks cannot get in the way of change, however. Becoming AI-first is a common goal in the enterprise, and those who pass on AI implementation are missing out.
"Some companies just say we're just not going to do AI, we're just going to keep doing things the old way," Borne said. "But I think those companies are going to lose in the long run, because the technology is racing ahead so fast that the companies that get it right will just win the market."