2022: Year of ChatGPT, What Is The Future?
Looking into the crystal ball
In the last few decades, the field of natural language processing (NLP) has made tremendous strides in recent years. From virtual assistants like Siri and Alexa to machine translation tools like Google Translate, NLP technology is becoming increasingly prevalent in our daily lives. But what does the future hold for this ever-evolving field excitement?
What makes my heart beat a bit faster is the increasing ability of machines to understand and generate human-like language. I use the word Understand as a placeholder. This is known as conversational AI (the now famous chatGPT), and it’s a key focus for many companies and researchers. The goal of a conversational AI is to create systems that can not only “understand” what we say but also “respond” in a way that is natural and appropriate for the context. Here again, understanding and responding in a grand schema are concepts that are yet to be clearly defined, so please take it with a grain of salt.
The Rise of Large Language Models
Let us dive deeper. What has caused this transition from a deterministic NLP to something of the sort that can only be described as the “generation” of the text? Close enough to be called understanding or responding. For that 2 chemicals needed to be mixed. Deep learning is one of them. Deep learning allows machines to “learn” from large amounts of data. It can be used to improve the accuracy and naturalness of language generation. For example, a conversational AI system might be trained on a large dataset of human conversations. What does learning mean? Well, in this context, identify the next word or sentence based on the context of the previous word or sentence or paragraph, etc., and then use this knowledge to generate appropriate responses in real-time.
The Large of the LLM comes from the tremendous amount of textual data being used to train these models. This data is also known as a common crawl in which crawling on the internet is used to gather textual data. One of the key advantages of these models is that they can be fine-tuned for a specific task or application. For example, a language model might be trained on a large dataset of medical records and then fine-tuned to generate summaries of patient visits. This allows the model to incorporate domain-specific knowledge, which can improve the accuracy and usefulness of the generated text.
A number of companies and organizations are working on large language models for conversational AI. OpenAI is a research organization that is focused on developing and promoting friendly AI. They have made significant contributions to the field of NLP, including the development of language models like GPT (Generative Pre-training Transformer) and GPT-2, GPT-3, and now GPT-4. These models are trained on massive amounts of data and are able to generate human-like text with impressive accuracy and naturalness. In addition to OpenAI, many established companies and startups have contributed to the field of AI:
- Google has made significant investments in NLP and has developed a number of language models BERT (Bidirectional Encoder Representations from Transformers) for use in various applications.
- Facebook has developed RoBERTa (Robustly Optimized BERT Approach), which is a variant of BERT that was developed to improve upon the original model’s performance.
- Microsoft: Microsoft has a long history of research and development in the field of NLP, and has developed a number of language models for use in various applications.
In addition to these large technology companies, there are also many smaller startups and research organizations that are working on NLP technology, including Hugging Face, Element AI, and DeepMind. These organizations are often at the forefront of the latest developments in the field and are helping to drive innovation in the field of conversational AI and NLP more broadly.
The work of OpenAI and other organizations is helping to advance the field of NLP in a number of exciting ways. Ultimately these steps are paving the way for the development of more sophisticated and human-like conversational AI systems of the future. There is no doubt that these companies will continue to invest in the development of models that may one day be universal..
A Conversation Is Not Limited To Words.
Another trend in NLP is the increasing use of multimodal inputs. According to the presentation by Jon Gillick from UC Berkeley, NLP could be done using written, spoken, sung, visual, tactile, and situated in real world. Processing a variety of input types, including audio, video, and even images, allows for a more natural and intuitive way of interacting with machines and opens up a whole new range of applications.
For example, a machine translation from spoken text to translation to any other language in real-time could make it much easier for people to communicate with one another across language barriers. Similarly, a virtual assistant that can process video input could be used to provide visual assistance or guidance or to recognize and respond to gestures and other nonverbal cues.
The multimodal models could be extended into various applications:
- Sentiment analysis could be extended to factor in thetone, facial expressions, and body language of a speaker to determine their emotional state.
- Speech recognition could be extended to emotion recognition (sarcasm) and translation or text generation etc. This could greatly improve the interpretation of the spoken language.
- Image recognition could be extended to identify and classify objects, people, and other elements in a scene and to create new images like DALL-e etc.
Overall, the use of multimodal inputs is a promising trend in NLP, and is likely to play a central role in the development of more sophisticated and human-like conversational AI systems in the future.
Commas Save Lives
One of the most exciting possibilities for the future of NLP is the potential for it to be used in a variety of different fields. From healthcare to education, there are countless ways that NLP technology could be used to improve our lives. For example, in healthcare, NLP could be used to analyze large amounts of patient data and identify patterns that might indicate a particular disease or condition. In education, NLP could be used to create personalized learning experiences, tailoring content and activities to the individual needs and abilities of each student.
This may lead to improvement of patient care, reduce costs, streamline processes, reduce the burden for physicians, clean the data, and standardize research! Currently, a few foci come to the surface where significant work has already been done -
- Medical record management NLP could be used to extract important information from medical records, such as diagnoses, medications, and test results. This could help to streamline the process of managing and organizing patient records and could also improve the accuracy and completeness of the information.
- Clinical decision support indicates a particular disease or condition.
- Patient monitoring to monitor patient data in real-time, alerting healthcare providers to potential issues or changes in condition. This could possibly help monitor patients for adverse events.
- Medical language translation NLP could be used to translate medical documents and other materials into different languages, making it easier for patients and healthcare providers to communicate and understand important information.
It will be safe to expect that this multi-trillion dollar industry could use some sweet power of AI to reduce waste. Let’s say that there is too much of an incentive not to use it!
Immerse Me In An Alternate Reality
Although Zuck has left a bad taste in my mouth, another area where NLP could have a big impact is in the realm of virtual and augmented reality. As these technologies become more advanced, it’s likely that we’ll see more and more NLP-powered virtual assistants and other applications. For example, a virtual assistant could be used to provide guidance and assistance as you navigate a virtual world or to translate spoken dialogue in real-time as you converse with virtual characters.
Although there is a buzz around the usage of Virtual Reality for education, I think that it will remain mainly for entertainment purposes. Movies, games, adult entertainment, and whatnot. NLP could be an additional dimension to monotonous gaming experiences. I can see AI being able to generate text or translate text, while generative AI creates Avatars of the person of interest and outputs a video of the avatar presenting the video to various audiences in their own language. NLP in AI could truly break the boundaries of nationality.
Even mortality has been challenged by the recreation of Avatars and songs by famous singers and performers like Michael Jackson. One can imagine reliving the memories of those times by creating a multimodal presentation of the person and maybe writing some new songs. The possibilities are truly unlimited and reasonably realistic.
Conclusion
Overall, the future of NLP looks bright. From conversational AI and multimodal inputs to a wide range of applications in different fields, there’s no doubt that NLP technology will continue to evolve and play a central role in our daily lives.
