50 Shades of Artificial Intelligence


Artificial intelligence [AI] is not new – some techniques are more than a decade old. Today, there are many shades of artificial intelligence, and we are able to apply these techniques to new data and workflows. Whilst new start-ups may have the advantage of easy access to software, they do not have the data stores that many large corporations have gathered over many years. Small AI software companies are seeking acquisition parties that have the right data set to extract the value from their development.

As machine learning is rapidly accelerating our understanding of AI, and how it may be applied to solving real world problems, and creating new opportunities, AI and machine learning are merging into a sphere of ‘machine intelligence’. However, as I see machine learning as a background task to evolving AI, I will for the purpose of this article refer simply to AI, but with the understanding that machine learning is an integral part of AI evolution.

Virtual agents, sometimes called bots, use conversational interfaces – some completely automated; others with a ‘human in the loop’ system where algorithms take machine-like subtasks and a human adds creativity or execution. It is these human interactions [keyboard or vocal] that continue to train the bot.

However, we must remember that it is the human perception of the experience that creates the value. Compare an online messaging chat with a support agent, compared to the frustrating navigation around an IVR [interactive voice response], and you will quickly recognise that AI machines have a long way to go in terms of taking over human tasks.

One of the difficulties in the adoption of AI in business is gaining acceptance of something that is difficult to explain in simple terms. Whilst many companies are reaching out to learn more about Artificial intelligence, and have crafted appropriate questions, their language tends to be more technical, rather in terms of solving a business problem.

Companies such as IBM and Google are reaching very wide across the Artificial intelligence landscape. Add to that governments, such as Canada, who has invested heavily in AI and continues to be central to the future of artificial intelligence. Groups like AICML to commercialize advanced research, the Artificial Learning Creative Destruction Lab

The future of AI will not be in generic software, but rather in AI software that solves a specific problem. For instance, at the macro level, AI company Orbital Insight is utilising satellite imagery with computer vision algorithms to determine the construction growth rate in China. At the micro level, computer vision can also detect subtle signs of PTSD, or remove tumours using autonomous surgical robots.

As companies continue to wrestle with the challenge of how to leverage AI, there are a few elements that provide insight into how AI may appear in the marketplace, along a continuum from raw data to a merging of man and machine as cyborgs:

Big datasets – the value of AI is maximised in big data sets; often extracting answers to questions previously remaining unanswered for many years. For those who do not have their own internal data sets, third-parties have spawned at a rate where data is becoming highly commoditised.

Niche insight – with wider access to commercial datasets, smart companies are now recognising the value in holding niche data close. This data solves very specific problems that are likely to provide a competitive edge. However, they must also own the algorithms to extract that insight – models that will likely not be commercially available for a smaller niche problem.

Data Alchemists – utilising a combination of the above two assets, data solution providers can enhance and enrich big data sets using niche algorithms and third-party data to reveal powerful insights. This has a high consulting component, as well as a technology and data science component.

Data Insight Brokers – provide insights from client and third-party data rather than their own data. These specialists analyse a wide array of data types and use cases, extracting new insights not previously available. The danger in this approach is that the thrill of the revelation may be short-lived if it does not relate to solving a specific problem, or creating a new profitable opportunity.

Workflow Enhancers – codifying lessons discovered in domain-specific data, and then integrating these new insights into existing workflow, to either automate outcomes or support human decision making. Whilst humans may lose expertise, these solutions will generally free up humans from more monotonous repetitive tasks, to apply their brain-power to ad hoc thinking and discovery.

Autonomous systems – from home help, personal assistants to self-driving cars. These systems are still rudimentary, but as intelligent machines are becoming more and more adept at self-learning, we can expect significant progress over the next five years.

Bots – are explicitly non-human and rely on human guidance to instruct them on what tasks to perform, and how to perform them. Tasks are less complex, for instance performing basic research, completing online transactions, monitoring and analysing data, or even task management.

Cyborgs – a blend of human and non-human intelligence to provide human-like experiences. They largely focus on complex tasks, such as customer service via real-time chat or meeting scheduling via email

Regardless of how AI will be applied, the primary challenge at present is concerning privacy and security, as we trust more and more of our data to machines.

Reference: Zilis, Shivon. ‘Machine Intelligence in the Real World’. techcrunch.com