Personalizing Open Domain Chatbots: How OLPORTAL Holds the Key

Open-domain chatbots distinguish themselves from other chatbots in terms of the diversity of functions they are capable of performing. The commercially used chatbots like Siri or Alexa have a very specific function to serve. However, unlike these, open domain chatbots are capable of performing different functions that are almost like a human. For instance, they are capable of appropriately responding to all kinds of comments from the users.

However, the most significant aspect of this technology is its ability to acquire a personality of its own, based on the training process. We will discuss some of the successes, failures, and solutions of this technology.

Facebook’s Blender chatbot has been able to dazzle the world with its revolutionary technology. It has been found to be most human-like compared to all other previous open domain chatbots from its closest competitors- Google and Microsoft. Google’s Meena had been one of the most human-like AI chatbots evolved before Blender came into the picture. It had a sensibleness and specificity average of 72% which is 86% for an average human being.

However, Facebook’s Blender chatbot has beaten Meena according to Facebook’s tests where the two systems were pitted against each other. 75% of the judges went for Blender in terms of a better conversational experience. And they claimed that 49% of the judges even stated that they would rather have a conversation with Blender than a similar human to human conversation. Yet the researchers stated in terms of development and training there is a long research trajectory that lies ahead of the team.

Moreover, Blender has not been made available to the public for use. This might have been driven by the experience from a similar open domain chatbot developed by Microsoft called Tay, which was launched in 2016. Tay was one of the most advanced chatbots and was pretty accurate. The company expected that the bot can be trained a larger set of skills by the users. But soon it turned out that the bot has turned racist, thanks to the users who were able to teach the bot racist statements and prejudices (oh the world we are living in!).

In the case of open domain AI systems therefore, the chances of exposure to spurious data is much higher compared to other specific function chatbots that are trained to ignore irrelevant inputs and requests. And in terms of developing a personality, it is often difficult to customize the personality according to the needs of individual users, as they are trained by the inputs of a large pull of users.

Is Decentralization the Answer?

Decentralization of the AI system by a technology like that of the blockchain can go a long way in handling both of these issues, along with creating various new business models. The OLPORTAL platform has created a system for pre-trained bots to be customized by the users and has also made chatbots available which are customized as specific fictional or non-fictional personalities. These can either be used as virtual assistants for the users or integrated into messaging applications to replicate these personalities, including that of the user which she can train while using it. And being specific and stored on a blockchain all the users of the chatbots will have a stake in preventing the bot from getting trained with spurious data.

Moreover, the OLPORTAL platform can be used for cost minimization in the development and publicizing of open-domain chatbots through its unique automated online market place driven by smart-contracts and algorithms. It can be used by users to create unique incentivizing mechanisms for the quicker development and training of the bots, unlike traditional methods.

To understand what we are building at OLPORTAL, please connect with us –

Website | Telegram | Twitter | Medium | Facebook | Reddit | Linkedin | Instagram | Kakao Talk