Beyond the fear: opportunities in AI

Whereas the capabilities of early artificial intelligence (AI) were confined to carrying out discrete tasks in specific and familiar contexts, those of the newest iteration – large language models (LLMs), such as GPT on which ChatGPT is built – are much advanced.

Trained on large quantities of data, able to understand a greater degree of nuance, and by generalising the knowledge they gain and transferring it from one problem or context to another (known as ‘transfer learning’)², their advanced abilities in reasoning and ability to learn from experience³ brings them much closer to the true definition of ‘intelligence’.

Such advancements, as characterised by the hype around ChatGPT, have captured people’s imagination by giving a glimpse into what’s possible. Whereas the previous versions of GPT (GPT-3) provided summation of information and scored in the bottom 10 per cent of test takers of a simulated bar exam, GPT-4 can suggest a course of action based on its analysis and scores in the top 10 per cent.⁴ Newer versions will naturally become even more powerful and useful.

The conversational nature of ChatGPT’s chatbot interface, which responds with human-like responses, and its ‘multimodal’ ability to accept both text and image inputs have further added to the perception of technological advancement in AI. This has thrust its potential impact on multiple areas of society further into the limelight, including the workplace which has left businesses asking how much of a threat it poses to the status quo and how they can use it to increase competitiveness.

Immediate application is likely to be in the form of supporting with administrative tasks and manipulating existing information into different formats (so called ‘generative AI’), such as planning and defining workscopes, producing reports, building websites and preparing bespoke client communications. Microsoft, for example, is previewing a digital assistant based on GPT-4 powered capabilities in its Office programs, which can summarise whiteboard content, draft sensitive emails and produce reports based on internal company data.⁵

The widespread deployment of the capabilities of large language models such as GPT-4 would be transformational. And, with billions of dollars of investment now moving into the development space,⁶ the underlying technology is quickly being turned into a multitude of use cases. These range from automating customer support, making real-time operational efficiencies, reviewing and editing (rather than just writing) code, and accelerating health diagnoses.⁷

A look at some of the investments made by SoftBank’s Vision Fund – the world's largest technology-focused investment fund – gives a glimpse into how a large investor is prioritising AI’s potential offerings.

• InterVenn is leveraging AI and deep machine learning technology to analyse proteins for cancer diagnoses and cures at early stages.
• eightfold.ai uses AI to help organisations futureproof their workforce by proactively assessing and predicting the impact on workforce size and skillsets of emerging and future trends.
• Observe.ai is using AI to detect customer sentiment and analyse interactions with customer service centres, and provide live guidance to agents to optimise the customer experience.

The early-stage nature of generative AI technology in particular means there are significantly more unknowns than assurances at this stage. These include intellectual property and data protection rights – how programs will store and use input data, and the legal ownership of that data and outputs – and how to address intentional and unintentional bias in the development and datasets used to train the models.

Reconciling these will influence the pace and shape of AI development. So too will societal acceptance of the technology, the ability to scale the necessary infrastructure and navigating complex regulatory landscapes. Regardless, the opportunities it presents for disruption are generational. Fields as diverse as software and digital media, medical science and healthcare, professional services, banking and the creative industries could be reoriented around its capabilities and the role(s) it comes to play within them.

It could also provide an opportunity to disrupt markets with dominant players. An example is the potential of AI-driven conversational search to become a paradigm shift in how users find information online. Added to the latest beta version of ChatGPT is the ability to search the web. Though currently restricted to select users and developers, it’s a capability that takes the challenge to Google and its long-standing dominance of the $US187b global search market.⁷

(Google has launched an AI-powered extension of its search engine – Bard – and Microsoft is integrating GPT-4 into its Bing search engine, but the hype around ChatGPT could see it make inroads, and any loss of share and advertising revenue would be costly for Google.)

The growing chorus of calls for government intervention to mitigate the risks (including from OpenAI’s CEO) is a sign that regulators appear to have been caught somewhat off guard by the growth and innovation of AI.

While oversight and control over AI platforms will increase, regulatory and policy approaches will need to adapt to reality as no nation or company will want to be left behind. The most likely outcome will be controls governing how companies market their AI capabilities, risk-based approaches to assessing and regulating AI, and industry-specific frameworks for the appropriate use of generative AI.⁸  

By providing much-needed guardrails, regulation can facilitate acceptance and stimulate the development of commercial use cases. Just as early AI and advances in machine learning have become part of life – from detecting financial fraud to making TV viewing recommendations, predicting machine maintenance, aiding drug discovery, and optimising ridesharing scheduling – this new era is part of the ongoing evolution of AI, albeit at an accelerated pace. Buckle up.