LLMs, Machine Learning, and Food! When AI Gets Tasty
Since the start of the COVID-19 pandemic in 2019, restaurants have seen an unprecedented increase in takeout and food delivery services. Online ordering and delivery platforms like UberEats and Doordash benefited greatly from this surge with most of them reporting record breaking numbers and profits. Uber Eats, for example, passed $8.3 billion in revenue in 2021 compared to $3.9 billion in revenue in 2020 and $1.4 billion in 2019. Although most restaurants and food services are relatively back to normal now, online ordering and delivery is still at higher rates than was projected pre-pandemic.
In a bid to reduce waiting times and increase efficiency, some restaurants and online ordering services are turning to voice recognition technology by incorporating language AI into various parts of their system including AI-based voice ordering systems. This would potentially help to reduce operational costs while making the ordering process more inclusive and accessible for people with disabilities who might find traditional methods challenging.
🍜 How it works
Most online ordering platforms that offer voice ordering do so by integrating existing voice assistants like Siri, Amazon Alexa, and Google assistant into their apps. UberEats, for example, integrates with Google allowing users to say a phrase that would open the UberEats app. The user can then place their order with the restaurant through voice while Google Assistant confirms the order. The entire process can be hands-free if the user wants it to be.
Other platforms, like Doordash, are building their own voice recognition systems to power voice ordering for their platforms. This typically involves using a number of algorithms and systems designed to create a personalized and convenient system for customers. These are;
Speech recognition: Speech recognition is typically used to understand and translate spoken language and voice commands into text by the system. For voice ordering, speech recognition would be instrumental in understanding voice orders and identifying relevant information.
Content-based filtering: Content-based filtering involves using the previous actions of a user or their explicit feedback to recommend other items similar to what the user likes. In this context, this would enable users to get recommendations unique to them. Recommender systems and ML algorithms can also be used for this purpose.
Natural Language Understanding (NLU): NLU is the aspect of Natural Language Processing that deals with the ability of a system to understand and interpret human language and not just individual words. This allows the voice ordering system to understand the fundamental meaning of sentences that may require some context and nuance that a machine would not have.
Natural Language Generation (NLG): NLG on the other hand is the process of generating output that resembles natural written or spoken language. This is used to generate natural responses for AI-based voice ordering.
🥐 Other uses of language AI in the food industry
While voice ordering on online delivery platforms is one of the more widely known uses of language AI in the food industry, there are other areas where this technology can be utilized to make various aspects of restauranting more efficient. One of them is by implementing voice recognition in drive-thru restaurants. Unlike voice ordering with online delivery platforms, this requires the employees at the drive-true use the voice recognition systems to streamline the process of accepting and dispatching an order. In one restaurant that implemented this system, customer complaints decreased while revenue increased by 10% to 20%.
Voice recognition can also be used to simplify the process of inventory control, allowing employees to document stock information directly into the inventory management system through voice, saving time and energy. Waiters can also communicate with other staff and compile orders more efficiently using voice recognition systems rather than having to take individual orders or use pen and paper.
🥑 Challenges of using language AI in the food industry
Introducing language AI to restaurants and apps is an expensive task that can mostly be afforded by large organizations. The implementation process alone requires a lot of data and trial-and-error that many smaller companies do not have the capacity for. There is also the possibility that users might find the voice ordering process unpleasant, especially users with thicker accents or who only speak other languages that are not English. Most of these challenges are common to every industry that uses AI and can be solved by more research into languages with less data as well as integrating existing voice technologies.
There are other more technical challenges that the food industry is facing. For one, it is one thing to build a voice ordering system and it is another to get high speed performances that are similar to one you would get with a human on the other end. Also, for restaurants with specially trademarked words or words that are not present in the dictionary (which is most of them) there needs to be lots of training data for these specific words since they won’t be present in most readily available datasets. This is necessary to make sure that the speech recognition model recognizes these words when a customer uses them.
AI has found a place in the booming food industry and recent advancements in language AI has produced various ways to streamline various processes in the industry, and serve customers better and more efficiently. With the increasing use of online delivery platforms, language AI in the form of voice ordering can serve to provide a more personalized and conversational experience for customers. There are even more possibilities that have not been explored yet and if the challenges are addressed, this could be the beginning of an exciting collaboration.