Examples of artificial intelligence are increasingly common these days. The element of deep learning is still in its infancy. It will provide increasingly great opportunities for knowledge managers, but its early iterations have raised some concerns. Softlink IC, like many others in the library technology space, is watching the evolution of deep learning with great interest.

Life Imitating Art

Popular culture introduced a world that included artificial intelligence (AI) and, what we now call deep learning, generations ago.

Orwell’s novel 1984, television’s Lost in Space and Star Trek, and movie productions such as 2001: A Space Odyssey, imagined AI. It is no longer the product of a writer’s imagination. It is part of our reality.

In the Beginning, There was Just Artificial Intelligence

Artificial intelligence uses related technologies to produce goods more efficiently and safely to perform repetitive tasks traditionally performed by humans. It has evolved from simply replicating physical movements, to speech recognition and learning.

Globally, industries around the globe work more efficiently, productively, and safely. Businesses and governments have already applied it to a variety of fields, including defense.

What Is Deep Learning?

Machine learning (ML) is a subset of AI. It focuses on learning from data to identify patterns and then make decisions with minimal intervention.

Deep learning applies some of the ML elements but requires vast quantities of existing data from which to learn. It is expensive, and as we are discovering, the current data is presenting issues that must be resolved.

Deep learning is better than humans at identifying patterns in data and analyzing images to predict outcomes. What is critical to its successful application is the quality of the data – its accuracy, the identification of underlying biases, and credibility of sources.

Concerns Regarding Deep Learning

Many believe there are significant concerns relating to, AI and deep learning that must be overcome.

They include:

• Humans (and bots) create data. Much of the data is subject to conscious or unconscious bias. The quality of the data is critical to the performance of AI and deep learning. That is still a way off.
• How to program “morals”. Morals can be viewed as objective and subjective. An example of this is killing. While there may be collective agreement that killing is morally wrong, there are cultural reasons why it may not be.
• Languages and their complexities. Take English as an example. It includes words that sound the same but are spelt differently, and with totally different meanings.
• Applying privacy and confidentiality requirements.
• A lack of transparency.
• The challenge of understanding abstract ideas and causal relationships reinforces the need for vast amounts of data.
• Humans write code and devise algorithms. Humans are fallible and algorithms are vulnerable to “spoofing”.
• Unforeseen circumstances.

There are fears that computers could develop “minds” of their own. Two years ago, Facebook was the subject of media attention due to an issue with two of its AI computers. It was reported the machines had ditched English as their mode of communication and invented their own language. The Tolkien of the AI world.

Algorithms will become exponentially more complex to deliver optimum results from deep learning. It will present us with enormous opportunities, but it should not be embraced without awareness of the significant downsides that could result from a lack of vigilance.

In an era of big data, there is no doubt our roles and workflows will change in ways determined by deep learning. For the information management sector, it will enable knowledge managers to more easily analyze data and deliver information that is more granular, to those who need it in their organizations.

Conclusion

AI and deep learning is here and evolving, but it is still early days. Suman Kanuganti quotes a Garry Kasparov comment that “We are in the beginning of MS-DOS and people think we are Windows 10”. Kanuganti himself stated, “AI realistically is like a 3-year-old child at this stage. When it works, it feels magical. It does some things well, but there’s still a long way to go.”

Most 3-year-olds do not always exhibit problem-free behaviour as they evolve. The same could be said of deep learning.