Cyberlife Executive Summary

Industry Overview

Multiple prominent figures in the Android development space have likened Androids to home computers. During the 1980s, computers reached a tipping point, and the technology improved enough so that it could be cheap and user-friendly for non-technical users. Nowadays, computers, phones, and tablets are inseparable from modern life thanks to the pioneers who first introduced them to the public.

The Android market is on the same precipice. Currently, Androids have found their biggest markets in customer service lines, disaster relief operations, and 'adult' uses. Several start-ups are attempting to penetrate the market of individual Android use, but none have been successful, deterring investors from this extremely lucrative and undersaturated market.

The most notable reason for these continued failures is the perceived lack of 'humanity' within Androids. Users often become uncomfortable with the 'robotic' behavior of the Androids, especially in regards to their body language and response to stimuli. The current way forward is for companies to create Androids that simulate humanity as closely as possible, while still providing a machine that will execute any orders given to it.

Technological Overview

Current Android production methods focus on starting from a machine base, using simple input-response methods to recreate human-like behaviour and decision-making within the Android's actions. These Androids are equipped with delicate — and often costly —  sensors to translate visual, auditory, and spatial data into machine-readable data. This data is then analyzed thoroughly by behavioral algorithms implanted into the Android, allowing the Android to respond with whatever the algorithm deems appropriate, an operation-heavy process. This will be referred to as the Machine-based, or M-base model.

Cyberlife has taken a different approach to modelling 'humanity' in Androids. Using data from several hundred brain scans, a blank 'template' of a brain is created inside software. Cyberlife then uses machine learning to 'reduce' the neurons that lead to emotion and ego, then modifies areas of the brain to allow goal-orientation that aligns with a user's commands. This will be referred to as the H-base model.

Strengths

The H-base model vastly succeeds the M-base model in several areas:

• The response time of the Android and the amount of processing power are reduced as the Android 'brain' does not need to convert sensory input to machine-readable data; the neural network adapts to its environment after only hours of testing, much like a human brain.
• The H-base avoids the 'uncanny valley' effect of current Androids, being able to mimic human emotions and gestures with ease.
• The M-base can, at times, get stuck in dialog or behavioral loops when entering an unfamiliar situation. Currently, this can only be broken with the help of a technician. This is especially relevant to disaster recovery, as the environment leads to many unexpected situations that can cripple M-base Androids. The H-base can be unsure of what to do in extraordinary situations, but will always be able to defer to humans in such circumstances. It can also learn and improve on itself, and will respond in an advanced manner when a similar situation arises.
• The H-base saves memory and processing power by being able to distinguish between a user issuing commands and irrelevant social scripts. This can also be applied to its environment; it has the ability to discard visual, auditory, and other data until it becomes relevant, e.g., only monitoring temperature sensors when a user’s body temperature becomes important data.
• The H-base can differ its response speed based on a user's tone. M-base models are often criticized for moving at an ‘uncannily’ efficient pace. Modifying the behavior by slowing movement down still frustrates users, perceiving the M-base as 'apathetic', moving at similar speeds when there is an emergency. The H-base will appear more urgent when its user appears more frantic or demanding, satisfying human psychology to make the Android appear more personable.

Weaknesses

Because of the novel approach taken by Cyberlife, H-base models will need more time for researching current defects:

• H-base Androids are currently less stable in certain areas, 'glitching' when exposed to certain stimuli in testing, where they will begin to disobey orders and mimic human distress. Heavier research and testing are needed to determine how to stabilize H-base models.
• H-base models are unable to communicate data to each other at the same speed as M-base models. H-base models are unable to condense the data shared down to the efficiency of M-base models, and will often mimic humans by talking, which is exceptionally slower than communicating via serial interface.
• The testing of H-base models, because of Cyberlife's desire to make them appear as human as possible, can be distressing to watch without the knowledge that H-base models are designed for human mimicry. Cyberlife will need to establish the idea of this mimicry to the public, especially key regulators of the market, to avoid misplaced empathy. Security measures are in place so that no testing documentation will be leaked.
• Cyberlife has been unable to prevent H-base models from mimicking screaming. The ability to cry has been successfully removed, as well as the ability to mimic heaving/throwing up, limping, and tending to 'wounds'. This is to avoid Androids from showing any undesirable human features. The ability to scream has been managed so as not to appear in routine situations, but currently, high-stress stimuli still result in screaming mimicry. Cutting access to vocal routines in high distress situations is a potential option, but may be potentially hazardous for Androids in disaster relief efforts.
• H-base models give limited access to fine-grain facial feature control. M-base androids can be fine-tuned to display a variety of facial emotions and are easily controlled. However, because of the unique nature of the H-base, emotions are displayed by ‘probing’ certain areas of neurons. This gives a much more realistic effect, but as a newly-developing technology, requires more research to achieve control over microexpressions. 

Research Overview

The Research and Development team is constantly fine-tuning testing based on user feedback from our alpha line. As an example from recent findings, studies indicate that Androids mimicking breathing increase familiarity with users. However, Androids panting from excessive physical activity or exertion distresses users, leading to some feeling disinclined to continue using the Android at states of high exertion.

Current research focuses on removing erratic responses when the Android chassis is damaged. The US Android Regulation and Safety Act outlines that Androids need to inform the user when they are encountering stimuli that will lead to partial or complete destruction of the Android. Cyberlife is in the stages of stress-testing H-base models, and scanning reveals the Android’s neural pathways are overburdened by damage to the chassis, overriding given warning protocols, and causing the Android to perform unregulated responses (screaming, shuddering, twitching, etc.).

Development has progressed so that this erratic reaction will only occur when the Android is near defunct. Because of the unique model of neural pathways the H-base contains, the Android will become more resistant to stress if it has previously become exposed to distressing stimuli. This ‘learning’ can be transmitted to other Androids artificially. Unfortunately, this also leads to decreased efficacy in Android’s goal-oriented software, where the Android will operate with excessive caution towards potentially distressing stimuli compared to control groups. Further research is needed in this area.

Roadmap

With backing from venture capital investors, Cyberlife Android hubs will first be opened in Detroit, then New York and Chicago. Androids will be available to rent on a subscription model to businesses, below the current minimum wage in the city.

Personal Androids will be sold at lower-than-market rate to outsell current companies. Once the company goes public, regulation issues can be addressed, and several interest groups can be created to push back against set limits. Android hubs can then expand into several major cities.

Androids with more impressive features, paired with higher prices, will be rolled out, and repairs for the previous line of Androids will slowly stall, allowing customers to switch to higher-priced models.

STRICTLY CONFIDENTIAL