Odyssey journey to AI
The journey to hyper personalisation.
Artificial Intelligence presents enormous opportunity to explore the world, identify patterns within it and place robots at the front line to do boring, complex or dangerous work. AI works by harvesting large amounts of data, evaluating patterns to be described using maths and pointing it in the direction of good. At Moroku we think that good involves using AI to help people thrive with their money.
Artifical Intelligence’s accuracy and usefullness is dependent on three factors:
- The data sets upon which it is based,
- The depth and learning ability of the algorithms
- The purpose and bias with which it is directed.
The first factor is foundational . Moroku CEO, Colin Weir, learnt this during his first attempt at building algorithms in the 1980s, during his post graduate research. The heart of his thesis was the development of a mathematical model to determine the economic impact of management decisions on the plantation forests of New Zealand. Past knowledge suggested that yield curves had a shape which was sigmoidal and could be described by a yield equation, with various coefficients describing the impact of yield (Y) over time (T). As more data was collected and coefficients analysed a new parameter, competitive index was found to greatly improve the accuracy of the predicitions.
During the 1980s research was conducted manually, adding and testing the utility of parameters. The key power of AI is to continue to add and test parameters to the algorithms as it learns the connection between inputs and outputs, such as with competitive index and yield. By adding and testing additional parameters, AI is able to create algorithms at speed that self adapt to fit data distributions.
ChatGPT and the AI development of creative works by scouring art works and creating new ones show us that there is much to be done in:
- the accuracy of engines
- their guidance and governance
- IP ownership rights of the original data sets upon which they learn
Artifical Intelligence, Machine Learning and Large Laungauge Models offer us many opportunities. As Stephen Hawkings, Noah Harrari and others warn, it also comes with some existential challenges. Most predominantly is the potential of the robots to take over from us , as they ignore our needs , those of the planet and begin optimising for something else; themsleves or worse. These concerns are not without due cause, requiring strong governance at an enterprise and political level.
ChatGPT was launched by OpenAI in November 2022 on top of OpenAI’s GPT-3 family of large language models (LLMs). Its capability to create the next generation of search engines, going beyond mere searching to interpretation and the consolidation of individual pieces of data (links) into consolidated thought pieces, has given Microosft its first real chance at competing with Google in web search, by offering a better user experience. With the democratisation of these techologies, AI has been released into:
- schools allowing students using AI to cheat in exams or turn in AI responses to homework
- HR departments for the creation of job applications,
- The web for the creation of blogs
- Call centres with Chat Bots replacing humans for improved efficiency.
One set of AI tools of particular relevance are large language models, or LLMs. These are deep learning algorithms that can recognize, summarize, translate, predict and generate text and other content based on knowledge gained from analysing massive datasets. Anywhere language is used, LLMs are useful: Translation, chatbots, search engines, composition of poems, songs and stories, writing software or moleculular and protein structures for new vaccines or treatments. Financial advisors can use them to describe financial markets, banks can use them for anomoly and fraud detection, legal teams can use them for paraphrasing and scribing.
Large data sets are fed into an algorithm using unsupervised learning, i.e. without explicit instructions on what to do with it. The model learns words, the relationships between and the concepts behind them. This knowledge can be used to predict and generate content.
AI has a lot to offer financial services, Moroku’s mission, the hyper-personalisation of customer financial journeys and the tailoring of unique digital products and services. As with the forestry example earlier, we begin with an initial cut of the algorithms, spread across a representative group of player leagues. These place the customer on the map and exist within the context of an individual bank or client of Moroku. Through a series of increments, more data is added through psychological archetyping and open banking to grow the data set, increase the granularity of the algorithms and the pattern exploration. Eventually the deep learning takes over the algorithm generation as players have algorithms refined just for them. This is how we define hyper personalisation, using AI, in the context of financial services.
At the heart of this journey towards a hyper personalised experience is the requirement for empathy, to go beyond logic, reason and patterns to unlock feelings and emotions. This remains a challenge for AI . We know we act based not on what we think but on how we feel. To improve its utility, AI must go beyond trawling the internet for previously defined or provided answers, to searching the very many answers available on the internet, synthesise their behaviours into a range of view points, qualify them as to their factual evidence, over lay them with some perspectives of the unknown and then apply the human factor. For those looking to ChapGPT to complete their exams, apply for a job, respond to a RFI or RFP, there remains a long way to go. For those looking to create connection there looks even further to go.
Artificial Creativity
In addition to content, images made by AI are suddenly all over forums on the internet where users share works they have generated from text prompts. The two most prominent of these systems are Dall-E by OpenAI and Stable Diffusion, created by Stability AI.
For these algorithms to work they have to copy prior works and then use it to train their AIs. This is the basic mechanics of all so-called generative AI—first, a company finds or creates a big enough set of data, then uses various algorithms to train software to generate specific text, images or code based on that data. The art generated using these methods threaten to bury the original works and any declaration or acceptance of IP ownership.
AI that excels in pattern recognition, enabling functions like recognizing family members’ faces in photo apps or flagging objectionable content before it reaches our social-media feeds has become common place. The algorithms that power generative AIs go one step further, to produce new content, not just recognize what already exists. Whilst AI has the potential to transform the production of all kinds of creative work including novels, marketing copy, news articles, video, illustration and code, maintaining IP attribution seems to be important.
AI is amazing. Beyond stripping cost out of customer service and creating funky pictures there is real potential to help customers thrive in the new digital world.