Review of the state of the art in autonomous artificial intelligence

2025-04-29 0 0 506.64KB 8 页 10玖币

侵权投诉

Vol.:(0123456789)

1 3

AI and Ethics

https://doi.org/10.1007/s43681-022-00176-2

ORIGINAL RESEARCH

Review ofthestate oftheart inautonomous artiﬁcial intelligence

PetarRadanliev1· DavidDeRoure1

Received: 5 March 2022 / Accepted: 16 May 2022

Abstract

This article presents a new design for autonomous artiﬁcial intelligence (AI), based on the state-of-the-art algorithms, and

describes a new autonomous AI system called ‘AutoAI’. The methodology is used to assemble the design founded on self-

improved algorithms that use new and emerging sources of data (NEFD). The objective of the article is to conceptualise the

design of a novel AutoAI algorithm. The conceptual approach is used to advance into building new and improved algorithms.

The article integrates and consolidates the ﬁndings from existing literature and advances the AutoAI design into (1) using

new and emerging sources of data for teaching and training AI algorithms and (2) enabling AI algorithms to use automated

tools for training new and improved algorithms. This approach is going beyond the state-of-the-art in AI algorithms and

suggests a design that enables autonomous algorithms to self-optimise and self-adapt, and on a higher level, be capable to

self-procreate.

Keywords Artiﬁcial intelligence· Autonomous systems· New and emerging forms of data· AI algorithms conceptual

design

1 Introduction

The topic of artiﬁcial intelligence (AI) becoming autono-

mous has been discussed since the 1960s. This article

reviews the current state-of-the-art in autonomous AI, with

a speciﬁc focus on data preparation, feature engineering and

automatic hyperparameter optimisation. Hence, this article

reviews and synthesises literature and knowledge from the

last decade and beyond. Using synthesised knowledge form

the reviewed studies, the article presents multiple algorithms

and tools in a conceptual design, to provide a new solution

for automating these problems. The conceptual design uses

existing AutoML techniques as the baseline for automating

and assembling AI algorithms, resulting with AutoAI design

superior to current AutoML. To build the AutoAI, modern

AI tools can be used in combination with new and emerging

forms of data (NEFD). The targets of automation are set to

autonomous: data preparation, feature engineering, hyper-

parameter optimisation, and model selection for pipeline

optimisation. The design is addressing the most important

challenge in the future development and application of novel

compact and more eﬃcient AI algorithms—namely, the

self-procreation of AI systems. The design consists of four

Phases, each Phase addressing a number of speciﬁc obstacles

(O). The design initiates with constructing training scenarios

in Phase 1 that will teach the AI algorithm to use the OSINT

(Big Data) for automatic ingestion of raw data from new and

emerging forms of miscellaneous data formats. In Phase 2

the iterative method is be used to assemble and integrate

autonomous feature selection and feature extraction—from

web sites, DNS records, and OSINT sources, building upon

the knowledge from autonomous data preparation and the

automated feature engineering. In Phase 3 speciﬁc scenarios

based on biological behaviours are designed for automatic

hyperparameter optimisation at scale. In Phase 4 a new auto-

mated model selection is designed for pipeline optimisation.

The design follows guidance from recent literature on fair-

ness and ethics in AI design [1].

2 Methodology

This article integrates the distant ﬁelds of mathematical,

computer and engineering sciences. The research has been

designed with an iterative methodology, and lessons learned

* Petar Radanliev

petar.radanliev@eng.ox.ac.uk

1 Department ofEngineering Sciences, University ofOxford,

OxfordOX13QG, England,UK

AI and Ethics

1 3

from each iteration are used in the design and control of the

next iteration cycle—referred to as ‘phases’. To reduce and

overcome the complexity of the iterative methodological

process, and to progress towards a better understanding of

the outcome of each iteration, a variety of complimentary

but diﬀerent techniques are used. For example, the article

intersects methodologies from engineering and computer

sciences, to resolve future concerns with autonomous pro-

cessing and analysing of real-time data from edge devices.

3 Design forautonomous AI—AutoAI

The design consists of 4 Phases. In Phase 1 automatic prepa-

ration and ingestion of raw OSINT data is synthesised for

the construction of training scenarios for automated feature

engineering, and to teach AI how to categorise and use (ana-

lyse) new and emerging forms of OSINT data. In Phase 2

the domain knowledge is applied to extract features from

raw data. Feature is considered as valid if the attributes or

properties of the feature are useful or if the characteristics

are helpful to the model. For the automation of feature engi-

neering two approaches are considered (1) multi-relational

decision tree learning, which is a supervised algorithm based

on decision tree and (2) Deep Feature Synthesis which is

available as an open-source library named Featuretools.1 In

Phase 3 selection algorithms [2] is used to identify hyper-

parameter values. The normal parameters are typically opti-

mised during training, but hyperparameters are generally

optimised manually and this task is associated with a model

designer. The automation scenario design with start with

building upon the knowledge from biological behaviours.

The particle swarm optimisation and evolutionary algo-

rithms can be used, both of which derive from biological

behaviours. The particle swarm optimisation emerges from

studies on biological communities’ interactions in individual

and social levels and evolutionary algorithms emerge from

studies on biological evolution. Second, the scenario design

can apply Bayesian optimisation, which is the most used

method for hyperparameter optimisation [3].

The combination of these methodological approaches

is considered for automatic hyperparameter optimisation,

but the concern is that edge devices are characterised by a

large number of data points, and new and emerging forms

of data are characterised by large conﬁguration space and

dimensionality. These factors in combination could create

a longer than adequate time requirement for ﬁnding the

optimal hyperparameters. Alternative method is a com-

bined algorithm selection and hyperparameter optimisa-

tion. The method selection includes testing for the most

eﬀective approach, starting from Bayesian optimisation,

Bandit Search, Evolutionary Algorithms, Hierarchical Task

Networks, Probabilistic Matrix Factorisation, Reinforce-

ment Learning and Monte Carlo Tree Search. In Phase 4

an automated pipeline optimisation is designed, comparable

to the Tree-based Pipeline Optimization Tool (TPOT) [4]

but for autonomous optimising feature pre-processors for

maximising classiﬁcation accuracy on a unsupervised clas-

siﬁcation task. The above analysis of current AI algorithms

is designed for low cost devices that contain substantively

larger memory than IoT sensors. In other words, this design

could work for a ‘Raspberry Pi’ device, but it won’t work

for a low memory / low computation power sensor. Given

this, the functionality of the proposed AutoAI needs to come

in perspective. The proposed design could be applied in the

metaverse, or in mobile phones, on edge devices with some

memory and power, or in the metaverse. The design cannot

be applied to sensors used to monitor water ﬂow under a

bridge, or the air pollution, or smoke detector sensors in

the forest.

3.1 Phase 1: automated data preparation

O1: develop open access autonomous data preparation

method (for digital healthcare data) from edge devices—

similar to the Oracle autonomous database,2 for autono-

mous ingestion of new and emerging forms of raw data,

e.g., OSINT (big data). The ﬁrst scientiﬁc milestone (M1)

is to build a new autonomous data preparation method that

can serve for training an AutoAI algorithm to: (1) become

self-driving by automating the data provisioning, tuning,

and scaling; (2) become self-securing by automating the

data protection and security; and (3) become self-repairing

by automating failure detection, failover, and repairment.

The new method design includes learning how to identify,

map and ignore patterns of data pollution (e.g., using direct

references to results obtained from OSINT queries) and

become more eﬃcient in autonomously building improved

algorithms. To ensure the success of the autonomous data

preparation method, a new scenario is constructed to teach

the algorithm how adversarial systems pollute the training

data and how to discard such data from the training scenar-

ios. While constructing the scenario, the search for training

data expands in new and emerging forms of data (NEFD),

e.g., open data—Open Data Institute,3 Elgin,4 DataViva5;

1 https:// www. featu retoo ls. com/.

2 https:// www. oracle. com/ auton omous- datab ase/? fbclid= IwAR2

Niqrm jTZ76 hj0gN a1gQU ixCLE WY4g4 tlvSc YK0fv lW6q8 HiXM-

QXeC2A.

3 https:// theodi. org/.

4 https:// www. elgin tech. com/.

5 http:// datav iva. info/ en/.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

AIandEthicshttps://doi.org/10.1007/s43681-022-00176-2ORIGINALRESEARCHReviewof the stateof the artin autonomousarticialintelligencePetar Radanliev1 ·David De Roure1Received:5March2022/Accepted:16May2022©TheAuthor(s)2022AbstractThisarticlepresentsanewdesignforautonomousarticialintelligence(AI),based...

展开>> 收起<<

Review of the state of the art in autonomous artificial intelligence.pdf

共8页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Review of the state of the art in autonomous artificial intelligence

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: