More and more efficiency is created every day through technological advancement in the areas of artificial intelligence (AI) and automation. Will the advent of machine learning prove a positive way forward or will it be too convenient by far?
When asked about technological innovations doing the work of humans or operating ‘autonomously’, which professions and advancements spring to mind? Line workers? Checkout operators? Autonomous vehicles?
Machine learning is the implementation of artificial intelligence (AI) to provide systems the capability to make independent decisions and experiences without the need to be programmed. This is done through computer programs accessing data and learning for ‘themselves’ (if you can call machines that).
Cognitive psychologist, computer scientist and Professor at the University of Toronto Geoffrey Hinton was asked could AI be ‘better than us’. A scary question given a blunt response: “I think we should think of AI as the intellectual equivalent of a backhoe. It will be much better than us at a lot of things. And it can be incredibly good – backhoes can save us a lot of digging. But of course, you can misuse it.” Brian Mac Namee, founding member of the Applied Intelligence Research Centre (AIRC) at the Dublin Institute of Technology (DIT) and lecturer at the School of Computer Science at University College Dublin (UCD), says that machine learning is highly valuable in breaking the monotony of mundane jobs.
“One big advantage is that machine learning offers the potential to automate dull, repetitive tasks. More importantly, perhaps, is that machine learning offers the opportunity to encode high-end expertise in what is not easily accessible in automated systems.”
However, he highlights two potential negatives to the new technology that could clock-up penalties going forward. Number one is the analysis of data trends: “Machine learning systems search for reliable patterns in data and so lean towards majority outcomes. More unusual (potentially interesting) cases can be ignored.
This means that machine learning systems can lead to poor decisions being made for these minority cases.”
Number two is human interaction becoming redundant due to the growth of machine learning: “Increased automation driven by machine learning does have major potential impact as human workers are replaced by automated systems. This has significant problems both in terms of what will replace these disrupted jobs and how societal structures like the tax system will cope with this disruption.”
AI’s machine-learning capabilities will see the big players in the financial services industry leap well ahead of new entrants to the market due to information already at their disposal.
Looking at the financial services industry, by 2025 thousands of jobs could potentially be lost to artificial intelligence. According to consulting firm Opimas, the worldwide asset management industry would be one of the biggest areas hit in eight years’ time with 90,000 jobs getting the chop due to machine capabilities. Also attributed to this are high fees from asset managers and human portfolio managers struggling to keep up with large datasets making informed investment decisions.
It is predicted that securities services (58,000), sales and trading (45,000), private banking and wealth management (24,000), trading and clearing venues (15,000), and investment banking (4,000) will also see major jobs losses globally.
However, on the upside for human interaction, technology, data and other tech-related fields will see an increase in jobs (27,000) by 2025 due to AI and machine learning.
“I think the tech industry is better protected than some others. Jobs mostly consisting of repetitive, simple decision-making are ripe for replacement by machine learning. So low-end data analysis is probably at most risk. Machine learning is also likely a major component of more complex systems that might replace human workers – for example, self-driving cars,”says Brian.
Investment and future implementation
Even in the early stages billions of dollars are being pumped into this industry. According to Co-founder and Managing Director of Opimas Axel Pierron, in his paper Artificial Intelligence in Capital Markets: The Next Operational Revolution, Opimas “expect[s] financial firms to spend more than US$1.5 billion on AI-related technologies and, by 2021, US$2.8 billion, representing an increase of 75%. This does not include M&A activity and investments in start-ups.” The paper also went on to say “Cognitive analytics and machine learning solutions will take the lion’s share of this investment.”
The focus from now could turn to ‘training’ machinery a lot more than humans. A cost saver but also an efficiency generator, AI’s machine-learning capabilities will see the big players in the financial services industry leap well ahead of new entrants to the market due to information already at their disposal.
Pierron went on to discuss the hardship newcomers to the financial markets will experience because of more established players getting ahead due to their data-collection facilities already in place. He explained that AI will “reinforce the business model of financial institutions” and the fact that banks need access to huge amounts of data to “efficiently train an AI system”, will mean the existing organisations will have a massive head start and advantage over new entrants to the market.
Is machine learning almost too efficient? It seems not in the initial stages. The talking heads in the field still see the value of the human even if thousands of jobs may fall by the wayside.
How can all this investment be put to use? The tech news site ZDNet gives five steps for a company to implement AI and machine learning:
- Learn how machine learning can help your organisation.
- Research competitors and what other businesses are doing.
- Choose a platform, i.e. companies offering their own services (Amazon).
- Create a strategy, i.e. will machine learning be used for decision-making purposes or business processes such as manufacturing?
- Create an implementation plan.
The final point segues nicely to Pierron’s explanation of the implementation of machine-learning capabilities. Speaking to Business & Finance, Pierron said, “There are four main approaches to machine learning:
“Reinforcement learning: this approach relies on having the machine conduct numerous trials. Based on the outcome of each attempt, it learns how to reach the maximum reward/positive outcome.
“Supervised learning: this method requires a human being to monitor the machine activity and inform it what the correct output for a specific input is. This is the most popular method of machine learning, as it requires less training.
“Unsupervised learning: this technique tries to mimic the learning process of a human being or animal that relies on observation to teach itself.
“Semi-unsupervised learning: this is quite self-explanatory.”
Organisations can choose from these four main approaches to see which best fits their activities and future goals if AI and machine learning is to be incorporated into their business activities.
Are we fully machine ready?
It never seems to matter if we are ready for the new technologies coming our way; we always seem to take it in our stride, adapt or just take it as is. But will machine learning really render a lot of what we do, and what people do for a living, redundant? Some industries may encounter disruptions but some, such as tech, may even see the opposite with new job opportunities.
“It’s theoretically possible [that machine learning could make certain jobs redundant], but, in practice, the widespread adoption of machine-learning solutions is likely to create even more jobs”, says Axel Pierron.
He also suggested that it would be difficult to eliminate ‘human error’.
“It can eliminate ‘human error’ in a defined use case, for a specific process. However, it’s unlikely to eliminate all ‘human errors’ in a business context. Finally, at a philosophical level, it’s a human creation, hence subject to its error.
“Once designed, the algorithm learns from a set of data that has and will be selected by human beings. I’m convinced that in the journey we’ll end up with a number of ‘artificial stupidities’.”
Brian Mac Namee takes a similar stance: “No. Machine-learning systems still have a lot of human input (by those data scientists that build and train machine-learning systems) and so are prone to biases that humans bring to the job. Similarly, machine-learning systems that utilise datasets are likely to encode all of the biases that are contained in those datasets.”
So, is machine learning almost too efficient? It seems not in the initial stages. The talking heads in the field still see the value of the human even if thousands of jobs may fall by the wayside. Remember, a loss of jobs always happens whenever a new technological advancement is introduced into society. Once we get used to it, new and innovative jobs will stem from it. Certain people in certain sectors will be affected but industries such as tech will be enhanced being at the forefront of AI and machine learning.