Real-World Problems & How Machine Learning comes into the picture

Written by Praveen Kumar for blogs at AI Technology & Systems

It’s an era of tremendous & electrically fast technological changes. Within a flick of an eye, the way we read, understand, share, communicate, & interact with technology changes. We are in the 4th industrial phase in which use cyber-physical systems to monitor, analyze, and automate business. With the help of Machine learning, we are now able to gain insights from historical data & make predictions to solve problems which were not possible in any of the industry-phases before. We can now store a tremendous amount of data as well as make large computations on it. Machine learning has come up with solutions from data which were not humanly possible. However, we still have a long way to go because we still have problems which we face real-world life community.

Anyone who is objectively searching for answers to AI/ML problems knows that we are still just tackling low hanging fruits. That makes a lot of sense since AI is vaguely defined and if at all you wish to make any progress in AI, then you should learn to simplify your objectives or goals. If you read media outlets, it may seem like researchers are sweeping the floor clean with deep learning (DL), solving ML problems one after the other leaving no stones unturned. In reality, they are not, researchers attack relatively simpler problems in the hope of collectively solving the bigger problems that is just how research works. For example, object detection is mostly about the recovery of 2D bounding boxes with some limitations on the various objects can undergo visually. Though, in reality, the problem is harder than that, object poses are not in 2D but they are in 3D and it is not trivial to extend a 2D bounding box based object detection algorithm to 3D precise object detection problem. However, simple datasets like MNIST are considered solved but that is a toy problem when it comes to real-world variation in digit appearance. ImageNet is also solved but it is still not as challenging as everyday vision problems human visual systems solve. So even when you read about algorithms beating humans on ImageNet, that is just a small part of the full picture. There is a lot of research work needed to be done in:

Computer Vision such as object detection: The so-called state-of-the-art object detection algorithms make silly errors a human would never make. In the vision, we have more problems like video understanding, segmentation, image-to-image translation and many more.

Language translation: We do have better translation algorithms thanks to long-short-term-memory (LSTM) and convolutional networks than say, 5 years ago but trust me, humans do far better in translation.

Adversarial examples: Even towards the end of 2018, we still haven’t solved the adversarial examples problem yet. ML models can still be fooled by adversarial attacks. It is interesting to note that work coming from Google AI shows that even time-limited humans

can be affected by some adversarial examples. But given enough time, humans are extremely robust to such tricks.

Unsupervised/reinforcement learning: ML algorithms are highly dependent on clean and labeled data but we know that a lot more information can be gathered by agents interacting and experimenting with the environment all by themselves. There has been a lot of work in the area of reinforcement learning (RL) especially from DeepMind and OpenAI teams.

Generalization: As others have pointed out here, ML models find it hard to generalize well to novel conditions that deviate from optimal performance conditions.

Zero/one/few-shot learning: Humans, animals, and insects can learn from very little information in a very complex setting. Trying to emulate such capabilities in machines has remained elusive even to this very day (the time of writing of this answer).

Some real-life scenarios:

1. Climate & Forest classification:

Studying & understanding patterns in weathers, climate changes, Atmospheric changes, lot of lives can be saved. Forest classification datasets are present on Kaggle as well as with the governing agencies, but yet, there have no steps taken forward to analyze Flora & Fauna classification based on regions & states. If this is done, we will be able to grow plants where we have scarcity and being able to make balances of all the types of plants in almost every region.

2. Accidents:

According to the WHO annual global road crash statistics, nearly 1.25M million people die in road crashes, on average 3287 deaths a day. An additional 20–50M are injured or disabled [ref: https://www.asirt.org/safe-travel/road-safety-facts/]. We need to build systems on streets which are able to detect any of these before happening and also build autonomous systems targeting accidents as one of the most important factors.

3. Privacy & Security:

Privacy & Security are the most important aspects of a person’s life. Personal Information must be kept to ones and ones only. No one has the right to acknowledge some else’s private data & and other information. Privacy & Security must be taken care of. Machine learning systems should be able to protect user’s data if anyone is trying to access by illegal means and authorization.

4. Ledgers & Online transactions: Blockchain is a technology that can help track transactions between users in a public ledger which was developed originally for Bitcoin. It is now used commercially for various applications such as tracking ownership documents, digital assets or voting rights. Combining ML Systems with Blockchain will not just eliminate 3rd party indulgence, but also will be able to provide a stamp of trust on each legal online transaction. It will also help in maintaining anonymity throughout the web and no breach of privacy & security can be exploited.

We are in one of the biggest transitions phases and with each passing day, old technologies are eliminated and new solutions are built which are smarter, better & secure. We are still a long way away from the target of making the world a better place but well on course of doing so. We have problems, but we also keep digging for the solution and it’s about to happen in some time sooner. Some of the problems mentioned above are already been under tremendous success but yet to be completed or deployed on a large scale, others might just need some more time & research. While we solve problems, it’s also important we also make efforts in adapting new methods, technological solutions as well as move ahead with the technology by supporting it to make it happen on a global scale.