What are the areas of artificial intelligence, which are impacting our lives? Artificial intelligence is broken down into a few strong areas that impact our way of life. Chat bots are much in the news but they play a very minuscule role. The most impactful role is that of natural language processing, that is, speech to text and text to speech. Most commonly known are Siri or the YouTube speech to text. These are all built on AI models stitched and contextualised for dedicated use case scenarios.
Then comes the most researched topic of machine learning: AI algorithms, with semantic network models and neural network models, are allowing certain platforms to understand a consumer or user behaviour and predict patterns of their behavioural usage and prompt recommendations based on that. This is what all intelligent platforms such as Twitter, Facebook, LinkedIn have built in.
The other area, which is starting to turn heads, is the space of deep learning where an AI algorithm works with human intervention. More specifically, it involves a programmer who creates a model that starts delving deeper. For example, if a team is tasked to analyse millions of digital pages or trillions of data sets to get some meaningful findings, they would most probably use the deep learning models of pattern match, fuzzy logic or neural network models. Visual search or today's big data analytics platforms are all adopting the deep learning models and methods.
With this background in place, let's come back to these chat bots.
Chat bots cannot operate just by themselves. First and foremost, they have to be fed a lot of data. This has to be followed through with rules specific or relevant to that business, logic applicable to that business model and then again loT data. The objective primarily is to account for the varied influences and complexities that we see it in daily lives that need to be interpreted into parameters for the systems to understand.
Once there is loT data, the efficiency levels slowly start to improve, as it has a lot of parameters to interact with. More importantly, if we take the CHAT out of the BOT, most bots are designed to act according to a very strong rule of thumb: Logic inserted by humans, rules inserted by humans and a lot of data.
Now, we come to a very critical factor of moderating the responses via the bots and all of that is clearly driven by human behaviour and judgement.
Now, consider a hypothetical scenario: one billion active users every day on Facebook. On an average, each user would have a minimum of four to five areas of interest. This means there are four-five billion areas of interest to monitor every day. Let's assume there are a lakh employees in Facebook across all verticals and let's say 50 per cent of these employees have to monitor data and upload trending details on a minute by minute basis. That would mean that 80,000 news items will be manually monitored, manually mapped to all buckets of preferences, evaluated and then posted as trending every minute.
In reality, what happens is vastly different: technology partners with human intervention. The list of topics you see is still personalised based on a number of factors, including pages you've liked, your location, the previous trending topics with which you've interacted, and what is trending across Facebook overall. But people involved take calls such as #lunch may be most discussed every noon but will not trend. While the technology takes care of data collation, people take the decisions based on relevant intelligence.
Over a time, AI will get better and it will render obsolete the job of human intervention. AI will develop the relevant intelligence that humans currently use. But that's still a while away. But before AI becomes part of the real world, corporations have to answer some tough questions:
> To what extent does one apply relevant intelligence in AI interventions?
> Just because something is trending does it really mean it is newsy? Like #lunch or #sleeping? Or repo rates crash - which may not be trending but is it a newsworthy item - with implications explained?
> Curate or edit?
> Will complete removal of human intervention actually remove biases? Or, will it expose information consumption to a herd mentality of most popular being most consumed?
To summarise, yes, we have made huge advancements in the field of AI and each level of progress has been truly exciting. However, we are yet to arrive at the stage where we can allow AI to completely operate on its own without human intervention. As things stand, human intervention and moderation will be the key to its adoption and evolution.
The author is CTA, omnichannel commerce, IoT & AI, SAP Global