Manifestation of AI

Manifestation of AI#

Automated lawnmowers and vacuum cleaners#

We have already taken into use simple robots to automatize everyday routine works, such as vacuum cleaning or lawn clipping. A robot performing these tasks requires the blocks shown in Table 1.

Table 1 The potential building blocks of the lawnmower or vacuum cleaner robot.#
Activity	Category	Example implementation
Perception	Position	GPS, camera, velocity sensors
	Collision	Proximity sensors
	Charge	Measure the voltage of the battery
Cognition	Location	GPS evalution, machine vision, dead reckoning
	Model	Map
	Planning	Search solutions to find optimal path
	Collision	Implement a tactics to deal with obstacles
	Charge	Estimate the state of the battery
Action	Drive	Control speed and steering
	Start/Stop	Start and stop operation.
	Charging	Go to charging station when needed

Self driving cars#

[2]

The self driving car would be much more complex than a lawnmower, but it has similar high level functions. The self driving car needs more sophisticated sensors and algorithms, because the risks in car driving are bigger due to the higher energy, larger world, and denser traffic, for example.

Table 2 Some potential building blocks of the self driving electric car.#
Activity	Category	Example implementation
Perception	Position	GPS, camera, velocity sensors
	Collision	Proximity sensors, radar, camera
	Charge	Measure the voltage of the battery
Cognition	Location	GPS evalution, machine vision, dead reckoning
	Model	Map, physical models
	Planning	Search solutions to find optimal path
	Collision	Implement a tactics to deal with obstacles
	Charge	Estimate the state of the battery
Action	Drive	Control speed and steering
	Start/Stop	Start and stop operation.
	Charging	Inform the driver of the charging needs, or go to charging station when needed

An interesting environment for developing AI for self driving cars is the CARLA simulator

Humanoid robots#

[3]

Humanoid robots can mimic humans. They walk like a person, and can make tricks like a person and therefore they seem intelligent. It is also evident that these robots needs a huge amount of sensors to control each of its’ joints to keep the balance and walk, navigate through complex environments performing human-like tasks. It seems quite apparent that they are machines performing complex tasks, so perhaps they are a manifestation of AI?

You may also take a look at this video, but do not take it too seriously. What do you think is true and what is fake? Is it Ok to bully a robot? Does an intelligent robot with artificial feelings deserve human rights?

https://www.youtube.com/watch?v=y3RIHnK0_NE&t=7s

Medical diagnosis#

_images/ComputerAidedDiagnosisBoom.png — Fig. 6 The utilization of AI for Computer Aided Diagnostis (CAD) in medical fields. [2]#

The most succesful areas for computer aided diagnostics according to Fujita [2] are:

Fundus photographs: Mass screening of eye dieseases such as glaucoma and retinopathy. Google published a paper which implies that even heart disesases could be recognized frmo fundus images using deep learning.
Dermatology imaging: Skin cancer diagnostics where AI has observerd to outperform a human specialist. However it is good to know that the main method for diagnosing skin cancer is still biopsy, not visual examination.
Pathological imaging, detection of lymph node metastasis.
Breast imaging, mammography
Chest imaging, classify normal and abnormal major thoracic diseases

The figure Fig. 7 shos the utilization of deep learning to segment the human organs based on the example segmentation made by human experts.

_images/Organs.png — Fig. 7 The automated segmentation of organs from Computer Tomograhpy (CT) images. [2]#

Deep learning has been particularly succesfull in medical diagnostics. The disadvantege of deep learning is that the decision algorithm is a “black box”, which structure cannot be easily examined and explained. It is very important particularly in medicine, to explain how the diagnosis has been made.

University of Vaasa has also a few publications and few PhD thesis about AI based medical diagnostics, such as [1, 10]

Earth observations#

Earth observation is the gathering of information about the physical, chemical, and biological systems of the planet via remote-sensing technologies, supplemented by Earth-surveying techniques, which encompasses the collection, analysis, and presentation of data. Earth observation is used to monitor and assess the status of and changes in natural and built environments.

Söderfjärden spectral image example#

A satellite image of Söderjärden region is shown in Fig. 8. The original satellite image acquired by European Sentinel-2 satellite consist of 13 different channels, representing different wavelenght. The figure below is a normal 3-channel RGB image constructed from 13 original channels.

_images/Soderfjarden_image.png — Fig. 8 An RGB visualisation of a satellite image of Söderjärden region, including training data for the classification of land use.#

The image also contains some example data shown with polygons, which represent certain types of land use, such as forest, water, dark soil, hay field and grass. This data can be seen as a true classification results performed by an expert. An AI algorithm can now be trained to repeat the same classification using all 13 channels of data from the original image. The trained classification algorithm can then be applied to the rest of the image as well, and classify all pixels. The result of this is shown in Fig. 9 image below.

_images/Soderfjarden_classified.png — Fig. 9 An RGB visualisation of the land use classification results from Söderjärden region.#

If the training data provided by the expert wouldn’t have been available, then supervised classification had not been possible, but unsupervised clustering could have been tried instead.

Discussion point

What could be achieved using unsupervised methods for analyzing satellite images when compared with supervised methods?

Further information about EO#

Read more about earth observations from ESA’s Observing the Earth page or try the Windy application to see some interesting facts about Earth based on both space-based and terrestial observations.

To get facts about the climate and pollution from all over the Globe would not be possible without space based Earth Observations (EO). Thefore EO is also very importanc for climate change mitigation activities.

Tackling with climate changes#

AI can be useful for many purposes when tackling with climate change, as listed by National Geography, in How artificial intelligence can tackle climate change. The climatologists have been using AI methods in predicting

The figure Fig. 10 shows which AI methods have been popular in prevention of climate change